Duc-
> First thank Jeff Brower very much for your clear
explanation
> to my answer why we have to use windows for signal
> processing before.
>
> For Nadav Haklai's question, i think it makes sense in some
> cases, for example when we want to encrypt the telephone
> voice, do not want other people can hear it, to my experince
> the way to do that is quite the same as Nadav expressed,
> we do some changes, of course the result is still in the
> voice band, and at the receiver side we undo the change to
> recover the original speech.
>
> Base on the way Jeff gave, I believe we can do that without
> affecting the voice quality. We still use tone in voice
> band, for telephone often 0-4K. There are 2 ways:
> 1) we send first only the sync tone for curtain interval of
> time and after that only the voice. At the receiver side,
> we base on the received signal, used some kind of filers,
> it is easy to know whether the coming signal is sync tone or
> voice(based spectral for example), if it is syn tone, do
> recover tone by VCO for example and get time-synchronized
> with the sender.
> 2) We send both syn tone and voice but with known SNR of syn
> tone and voice, assume the transmission line is good
> enough, we can recover the original speech and extract the
> tone by spectral subtraction (the simplest) and do the same
> as the 1)
>
> Above is what I think, if there is any thing wrong please
> feel free to tell me
These methods are not a good idea to implement for the same reason that
telephone companies never combine signaling
and voice at the same time: voice quality is unacceptable. Even going the
other way -- adding voice to signaling --
is a notoriously difficult problem (you can search on "talk off" to study
this).
To convey some amount of non-speech data in-band, you might look at audio
watermarking. Watermarking techniques
"spectrally spread" small amounts of data (such as coypright info), and mix with
audio. The idea is the data is added
as very low level white noise and is (hopefully) imperceptible to listeners.
But I don't think the time precision in
this method would be good enough for the exact frame synchronization that Nadav
needs. Also watermarking would
significantly increase the DSP resources required (MIPS and memory).
-Jeff
> --- On Tue, 12/1/09, Jeff Brower
wrote:
>
> From: Jeff Brower
> Subject: Re: [audiodsp] Audio signal sync
> To: "Nadav Haklai"
> Cc: a...
> Date: Tuesday, December 1, 2009, 9:52 AM
>
>
>
> Nadav-
>
>> That is correct. For now, the project is intended for some sort of
>
>> demonstration.
>
>> So the telephone line noise is irrelevant (i'll be only using 1 meter
std
>
>> audio line).
>
>> Although, i was hoping to find a way to sync those two systems without any
>
>> other link besides the audio line.
>
>> I guessed that syncing framed audio signals between systems is a common
>
>> problem and there must be a simple solution (using special sync frame
which
>
>> can be played and detected, adding FSK to the unused bw etc)
>
> I don't want to sound negative, as you're trying hard and seem to be
persistent. But we have to apply rational
>
> thinking, not wishful thinking. I would ask you this: if you think this is a
common method, then what audio systems
>
> can you name that do this?
>
> The basic problem you're facing: if you modify your audio signal "in
band", then you will significantly affect the
>
> quality of what people hear. Whatever modification you would make needs
substantial energy to be detected reliably
>
> and accurately enough to provide precise frame synchronization -- listeners
*will* hear it. But you haven't specified
>
> what quality you need and what degradation you're willing to accept.
>
> RF communication systems often require precise synchronization. RF methods
typically use bandwidth outside of the
>
> audio or data signal. For example, if you sample at 16 kHz but only transmit
telephony quality audio (i.e. speech)
>
> limited to 4 kHz bandwidth, then you might use the upper 4 kHz for "out of
band" information. You could send a 7 kHz
>
> tone short burst every frame, detect that and use its center, or peak power,
as the frame marker. In your
>
> receive-side DSP code you would apply a sharp LPF to remove this from the
audio signal.
>
> -Jeff
>
>> On Tue, Dec 1, 2009 at 12:39 AM, Jaime Andres Aranguren Cardona <
>
>> jaime_aranguren@ yahoo.com> wrote:
>
>>>> Hi all,
>
>>>>> Although Jeff's point is a certain one, maybe the project is kind of
demo,
>
>>> and needs some sort of link (say a DSP's serial port) to demonstrate
that
>
>>> the encoding and decoding is working on the DSP sides, and the link is
out
>
>>> of the question. The issue here is that the link is what is causing the
>
>>> trouble. In other words, the problem is to get to DSPs to talk to each
other
>
>>> through some link, that link can be a serial port (of the kind where
people
>
>>> usually attach ADCs, DACs, Codecs) or an SPI peripheral on the DSPs or
TWI,
>
>>> UARTs, whatever. And if this is the case, usually the HRM (Hardware
>
>>> Reference Manual) is all what the OP (original poster) needs.
>
>>>>> Maybe I am 100% lost here, but this is how I interpreted the question.
>
>>>>> Regards,
>
>>>>>>> Jaime Andrés Aranguren Cardona
>
>>> jaime.aranguren@ ieee.org
>
>>> jaime.aranguren@ computer. org
>
>>>>>>> ------------ --------- ---------
>
>>> *Von:* Jeff Brower >> *An:* Nadav >> *CC:* audiodsp@yahoogroup s.com
>
>>> *Gesendet:* Montag, den 30. November 2009, 19:00:55 Uhr
>
>>> *Betreff:* Re: [audiodsp] Audio signal sync
>
>>>>>>>>> Nadav-
>
>>>>> > I'm developing a real-time audio processing project which
>
>>> > consists of 2 DSP separated systems.
>
>>> > One system is encoder and other decoder.
>
>>> > between the systems there is a regular audio line.
>
>>> > the encoder samples the input audio data (8/16/24 bits,
>
>>> > 16Khz sampling rate) separates it into frames and
>
>>> > process each frame individually. The processed data is
>
>>> > played by the D/A to the audio line and is the input to
>
>>> > the decoder system.
>
>>> > the decoder system samples the data and decode the data.
>
>>> >>> > My problem is that the data is being processed in separate
>
>>> > frames and the decoder must know when each frame begin in
>
>>> > order to process it correctly.
>
>>> > How can i sync those 2 system using only the audio signal
>
>>> > line which carries the actual data?
>
>>> > Is it possible some data bits to the data somehow?
>
>>> > Can i use special frames to sync?
>
>>>>> Something is not making sense here. Normally two systems doing audio
>
>>> encode/decode communicate via digital transport
>
>>> -- that's the whole point of audio compression, right? It should like
this:
>
>>>>> _______ Dig I/O _______
>
>>> Audio <--> | codec | <-------> | codec | <--> Audio
>
>>> |_______| |_______|
>
>>>>> where digital I/O is network (e.g. VoIP), RF (e.g. GSM, WiFi,
satellite,
>
>>> etc), serial port (e.g. RS-422), etc.
>
>>>>> If you're connecting two systems via audio, then why do you need
>
>>> compression?
>
>>>>> -Jeff
Audio signal sync
Started by ●November 30, 2009
Reply by ●December 3, 20092009-12-03
Reply by ●December 3, 20092009-12-03
Jeff Brower,
The application I wanted to express is like this, you and I want to communicate safely through the public telephone service, we dont want any other can eavesdrop what we exchange, and of course we dont want the use some special service from telephone providers, we just design 2 devices (phones) as the normal ones added some features help us to do that.
About quality of voice, for the first way I said, syn tone (signaling) is just sent the starting time for synchronization purpose, after that only voice is sent and the tone is removed by device so it is ok for voice quality but it comes with a bit delay. For the second method, if we mix speech and syn tone at a given SNR we can recover speech by subtracting the syn tone at the received side, and the quality is not much less than the original I believe
Audio watermarking technique is new concept to me, it seems interesting and I have just taken a look, but it sounds just suitable for digital audio media like audio files,... for transmitting signal like telephone line, I guess it will have some problems like tramission quality, noise and for spread spectrum method, how to get the reference (original speech) to get back watermarking data from the processed signal.
hope can get some of your comments, very nice to discuss with u
--- On Thu, 12/3/09, Jeff Brower wrote:
From: Jeff Brower
Subject: Re: [audiodsp] Audio signal sync
To: "Duc Nguyen Anh"
Cc: a...
Date: Thursday, December 3, 2009, 5:52 AM
Duc-
> First thank Jeff Brower very much for your clear explanation
> to my answer why we have to use windows for signal
> processing before.
>
> For Nadav Haklai's question, i think it makes sense in some
> cases, for example when we want to encrypt the telephone
> voice, do not want other people can hear it, to my experince
> the way to do that is quite the same as Nadav expressed,
> we do some changes, of course the result is still in the
> voice band, and at the receiver side we undo the change to
> recover the original speech.
>
> Base on the way Jeff gave, I believe we can do that without
> affecting the voice quality. We still use tone in voice
> band, for telephone often 0-4K. There are 2 ways:
> 1) we send first only the sync tone for curtain interval of
> time and after that only the voice. At the receiver side,
> we base on the received signal, used some kind of filers,
> it is easy to know whether the coming signal is sync tone or
> voice(based spectral for example), if it is syn tone, do
> recover tone by VCO for example and get time-synchronized
> with the sender.
> 2) We send both syn tone and voice but with known SNR of syn
> tone and voice, assume the transmission line is good
> enough, we can recover the original speech and extract the
> tone by spectral subtraction (the simplest) and do the same
> as the 1)
>
> Above is what I think, if there is any thing wrong please
> feel free to tell me
These methods are not a good idea to implement for the same reason that telephone companies never combine signaling
and voice at the same time: voice quality is unacceptable. Even going the other way -- adding voice to signaling --
is a notoriously difficult problem (you can search on "talk off" to study this).
To convey some amount of non-speech data in-band, you might look at audio watermarking. Watermarking techniques
"spectrally spread" small amounts of data (such as coypright info), and mix with audio. The idea is the data is added
as very low level white noise and is (hopefully) imperceptible to listeners. But I don't think the time precision in
this method would be good enough for the exact frame synchronization that Nadav needs. Also watermarking would
significantly increase the DSP resources required (MIPS and memory).
-Jeff
> --- On Tue, 12/1/09, Jeff Brower wrote:
>
> From: Jeff Brower
> Subject: Re: [audiodsp] Audio signal sync
> To: "Nadav Haklai"
> Cc: audiodsp@yahoogroup s.com
> Date: Tuesday, December 1, 2009, 9:52 AM
>
>
>
>
>
>
>
> Â
>
>
>
>
>
>
>
>
>
> Nadav-
>
>
>
>> That is correct. For now, the project is intended for some sort of
>
>> demonstration.
>
>> So the telephone line noise is irrelevant (i'll be only using 1 meter std
>
>> audio line).
>
>> Although, i was hoping to find a way to sync those two systems without any
>
>> other link besides the audio line.
>
>> I guessed that syncing framed audio signals between systems is a common
>
>> problem and there must be a simple solution (using special sync frame which
>
>> can be played and detected, adding FSK to the unused bw etc)
>
>
>
> I don't want to sound negative, as you're trying hard and seem to be persistent. But we have to apply rational
>
> thinking, not wishful thinking. I would ask you this: if you think this is a common method, then what audio systems
>
> can you name that do this?
>
>
>
> The basic problem you're facing: if you modify your audio signal "in band", then you will significantly affect the
>
> quality of what people hear. Whatever modification you would make needs substantial energy to be detected reliably
>
> and accurately enough to provide precise frame synchronization -- listeners *will* hear it. But you haven't specified
>
> what quality you need and what degradation you're willing to accept.
>
>
>
> RF communication systems often require precise synchronization. RF methods typically use bandwidth outside of the
>
> audio or data signal. For example, if you sample at 16 kHz but only transmit telephony quality audio (i.e. speech)
>
> limited to 4 kHz bandwidth, then you might use the upper 4 kHz for "out of band" information. You could send a 7 kHz
>
> tone short burst every frame, detect that and use its center, or peak power, as the frame marker. In your
>
> receive-side DSP code you would apply a sharp LPF to remove this from the audio signal.
>
>
>
> -Jeff
>
>
>
>> On Tue, Dec 1, 2009 at 12:39 AM, Jaime Andres Aranguren Cardona <
>
>> jaime_aranguren@ yahoo.com> wrote:
>
>>
>
>>> Hi all,
>
>>>
>
>>> Although Jeff's point is a certain one, maybe the project is kind of demo,
>
>>> and needs some sort of link (say a DSP's serial port) to demonstrate that
>
>>> the encoding and decoding is working on the DSP sides, and the link is out
>
>>> of the question. The issue here is that the link is what is causing the
>
>>> trouble. In other words, the problem is to get to DSPs to talk to each other
>
>>> through some link, that link can be a serial port (of the kind where people
>
>>> usually attach ADCs, DACs, Codecs) or an SPI peripheral on the DSPs or TWI,
>
>>> UARTs, whatever. And if this is the case, usually the HRM (Hardware
>
>>> Reference Manual) is all what the OP (original poster) needs.
>
>>>
>
>>> Maybe I am 100% lost here, but this is how I interpreted the question.
>
>>>
>
>>> Regards,
>
>>>
>
>>>
>
>>> Jaime Andrés Aranguren Cardona
>
>>> jaime.aranguren@ ieee.org
>
>>> jaime.aranguren@ computer. org
>
>>>
>
>>>
>
>>> ------------ --------- ---------
>
>>> *Von:* Jeff Brower
>
>>> *An:* Nadav
>
>>> *CC:* audiodsp@yahoogroup s.com
>
>>> *Gesendet:* Montag, den 30. November 2009, 19:00:55 Uhr
>
>>> *Betreff:* Re: [audiodsp] Audio signal sync
>
>>>
>
>>>
>
>>>
>
>>> Nadav-
>
>>>
>
>>> > I'm developing a real-time audio processing project which
>
>>> > consists of 2 DSP separated systems.
>
>>> > One system is encoder and other decoder.
>
>>> > between the systems there is a regular audio line.
>
>>> > the encoder samples the input audio data (8/16/24 bits,
>
>>> > 16Khz sampling rate) separates it into frames and
>
>>> > process each frame individually. The processed data is
>
>>> > played by the D/A to the audio line and is the input to
>
>>> > the decoder system.
>
>>> > the decoder system samples the data and decode the data.
>
>>> >
>
>>> > My problem is that the data is being processed in separate
>
>>> > frames and the decoder must know when each frame begin in
>
>>> > order to process it correctly.
>
>>> > How can i sync those 2 system using only the audio signal
>
>>> > line which carries the actual data?
>
>>> > Is it possible some data bits to the data somehow?
>
>>> > Can i use special frames to sync?
>
>>>
>
>>> Something is not making sense here. Normally two systems doing audio
>
>>> encode/decode communicate via digital transport
>
>>> -- that's the whole point of audio compression, right? It should like this:
>
>>>
>
>>> _______ Dig I/O _______
>
>>> Audio <--> | codec | <-------> | codec | <--> Audio
>
>>> |_______| |_______|
>
>>>
>
>>> where digital I/O is network (e.g. VoIP), RF (e.g. GSM, WiFi, satellite,
>
>>> etc), serial port (e.g. RS-422), etc.
>
>>>
>
>>> If you're connecting two systems via audio, then why do you need
>
>>> compression?
>
>>>
>
>>> -Jeff
_____________________________________
The application I wanted to express is like this, you and I want to communicate safely through the public telephone service, we dont want any other can eavesdrop what we exchange, and of course we dont want the use some special service from telephone providers, we just design 2 devices (phones) as the normal ones added some features help us to do that.
About quality of voice, for the first way I said, syn tone (signaling) is just sent the starting time for synchronization purpose, after that only voice is sent and the tone is removed by device so it is ok for voice quality but it comes with a bit delay. For the second method, if we mix speech and syn tone at a given SNR we can recover speech by subtracting the syn tone at the received side, and the quality is not much less than the original I believe
Audio watermarking technique is new concept to me, it seems interesting and I have just taken a look, but it sounds just suitable for digital audio media like audio files,... for transmitting signal like telephone line, I guess it will have some problems like tramission quality, noise and for spread spectrum method, how to get the reference (original speech) to get back watermarking data from the processed signal.
hope can get some of your comments, very nice to discuss with u
--- On Thu, 12/3/09, Jeff Brower wrote:
From: Jeff Brower
Subject: Re: [audiodsp] Audio signal sync
To: "Duc Nguyen Anh"
Cc: a...
Date: Thursday, December 3, 2009, 5:52 AM
Duc-
> First thank Jeff Brower very much for your clear explanation
> to my answer why we have to use windows for signal
> processing before.
>
> For Nadav Haklai's question, i think it makes sense in some
> cases, for example when we want to encrypt the telephone
> voice, do not want other people can hear it, to my experince
> the way to do that is quite the same as Nadav expressed,
> we do some changes, of course the result is still in the
> voice band, and at the receiver side we undo the change to
> recover the original speech.
>
> Base on the way Jeff gave, I believe we can do that without
> affecting the voice quality. We still use tone in voice
> band, for telephone often 0-4K. There are 2 ways:
> 1) we send first only the sync tone for curtain interval of
> time and after that only the voice. At the receiver side,
> we base on the received signal, used some kind of filers,
> it is easy to know whether the coming signal is sync tone or
> voice(based spectral for example), if it is syn tone, do
> recover tone by VCO for example and get time-synchronized
> with the sender.
> 2) We send both syn tone and voice but with known SNR of syn
> tone and voice, assume the transmission line is good
> enough, we can recover the original speech and extract the
> tone by spectral subtraction (the simplest) and do the same
> as the 1)
>
> Above is what I think, if there is any thing wrong please
> feel free to tell me
These methods are not a good idea to implement for the same reason that telephone companies never combine signaling
and voice at the same time: voice quality is unacceptable. Even going the other way -- adding voice to signaling --
is a notoriously difficult problem (you can search on "talk off" to study this).
To convey some amount of non-speech data in-band, you might look at audio watermarking. Watermarking techniques
"spectrally spread" small amounts of data (such as coypright info), and mix with audio. The idea is the data is added
as very low level white noise and is (hopefully) imperceptible to listeners. But I don't think the time precision in
this method would be good enough for the exact frame synchronization that Nadav needs. Also watermarking would
significantly increase the DSP resources required (MIPS and memory).
-Jeff
> --- On Tue, 12/1/09, Jeff Brower wrote:
>
> From: Jeff Brower
> Subject: Re: [audiodsp] Audio signal sync
> To: "Nadav Haklai"
> Cc: audiodsp@yahoogroup s.com
> Date: Tuesday, December 1, 2009, 9:52 AM
>
>
>
>
>
>
>
> Â
>
>
>
>
>
>
>
>
>
> Nadav-
>
>
>
>> That is correct. For now, the project is intended for some sort of
>
>> demonstration.
>
>> So the telephone line noise is irrelevant (i'll be only using 1 meter std
>
>> audio line).
>
>> Although, i was hoping to find a way to sync those two systems without any
>
>> other link besides the audio line.
>
>> I guessed that syncing framed audio signals between systems is a common
>
>> problem and there must be a simple solution (using special sync frame which
>
>> can be played and detected, adding FSK to the unused bw etc)
>
>
>
> I don't want to sound negative, as you're trying hard and seem to be persistent. But we have to apply rational
>
> thinking, not wishful thinking. I would ask you this: if you think this is a common method, then what audio systems
>
> can you name that do this?
>
>
>
> The basic problem you're facing: if you modify your audio signal "in band", then you will significantly affect the
>
> quality of what people hear. Whatever modification you would make needs substantial energy to be detected reliably
>
> and accurately enough to provide precise frame synchronization -- listeners *will* hear it. But you haven't specified
>
> what quality you need and what degradation you're willing to accept.
>
>
>
> RF communication systems often require precise synchronization. RF methods typically use bandwidth outside of the
>
> audio or data signal. For example, if you sample at 16 kHz but only transmit telephony quality audio (i.e. speech)
>
> limited to 4 kHz bandwidth, then you might use the upper 4 kHz for "out of band" information. You could send a 7 kHz
>
> tone short burst every frame, detect that and use its center, or peak power, as the frame marker. In your
>
> receive-side DSP code you would apply a sharp LPF to remove this from the audio signal.
>
>
>
> -Jeff
>
>
>
>> On Tue, Dec 1, 2009 at 12:39 AM, Jaime Andres Aranguren Cardona <
>
>> jaime_aranguren@ yahoo.com> wrote:
>
>>
>
>>> Hi all,
>
>>>
>
>>> Although Jeff's point is a certain one, maybe the project is kind of demo,
>
>>> and needs some sort of link (say a DSP's serial port) to demonstrate that
>
>>> the encoding and decoding is working on the DSP sides, and the link is out
>
>>> of the question. The issue here is that the link is what is causing the
>
>>> trouble. In other words, the problem is to get to DSPs to talk to each other
>
>>> through some link, that link can be a serial port (of the kind where people
>
>>> usually attach ADCs, DACs, Codecs) or an SPI peripheral on the DSPs or TWI,
>
>>> UARTs, whatever. And if this is the case, usually the HRM (Hardware
>
>>> Reference Manual) is all what the OP (original poster) needs.
>
>>>
>
>>> Maybe I am 100% lost here, but this is how I interpreted the question.
>
>>>
>
>>> Regards,
>
>>>
>
>>>
>
>>> Jaime Andrés Aranguren Cardona
>
>>> jaime.aranguren@ ieee.org
>
>>> jaime.aranguren@ computer. org
>
>>>
>
>>>
>
>>> ------------ --------- ---------
>
>>> *Von:* Jeff Brower
>
>>> *An:* Nadav
>
>>> *CC:* audiodsp@yahoogroup s.com
>
>>> *Gesendet:* Montag, den 30. November 2009, 19:00:55 Uhr
>
>>> *Betreff:* Re: [audiodsp] Audio signal sync
>
>>>
>
>>>
>
>>>
>
>>> Nadav-
>
>>>
>
>>> > I'm developing a real-time audio processing project which
>
>>> > consists of 2 DSP separated systems.
>
>>> > One system is encoder and other decoder.
>
>>> > between the systems there is a regular audio line.
>
>>> > the encoder samples the input audio data (8/16/24 bits,
>
>>> > 16Khz sampling rate) separates it into frames and
>
>>> > process each frame individually. The processed data is
>
>>> > played by the D/A to the audio line and is the input to
>
>>> > the decoder system.
>
>>> > the decoder system samples the data and decode the data.
>
>>> >
>
>>> > My problem is that the data is being processed in separate
>
>>> > frames and the decoder must know when each frame begin in
>
>>> > order to process it correctly.
>
>>> > How can i sync those 2 system using only the audio signal
>
>>> > line which carries the actual data?
>
>>> > Is it possible some data bits to the data somehow?
>
>>> > Can i use special frames to sync?
>
>>>
>
>>> Something is not making sense here. Normally two systems doing audio
>
>>> encode/decode communicate via digital transport
>
>>> -- that's the whole point of audio compression, right? It should like this:
>
>>>
>
>>> _______ Dig I/O _______
>
>>> Audio <--> | codec | <-------> | codec | <--> Audio
>
>>> |_______| |_______|
>
>>>
>
>>> where digital I/O is network (e.g. VoIP), RF (e.g. GSM, WiFi, satellite,
>
>>> etc), serial port (e.g. RS-422), etc.
>
>>>
>
>>> If you're connecting two systems via audio, then why do you need
>
>>> compression?
>
>>>
>
>>> -Jeff
_____________________________________
Reply by ●December 3, 20092009-12-03
Duc-
> The application I wanted to express is like this, you and I
> want to communicate safely through the public telephone
> service, we dont want any other can eavesdrop what we
> exchange, and of course we dont want the use some special
> service from telephone providers, we just design 2 devices
> (phones) as the normal ones added some features help us to
> do that.
>
> About quality of voice, for the first way I said, syn tone
> (signaling) is just sent the starting time for
> synchronization purpose, after that only voice is sent and
> the tone is removed by device so it is ok for voice quality
> but it comes with a bit delay.
What purpose is served by a "one time" sync tone? Nadav wants to do frame-by-frame synchronization. That would mean
a sync tone every 10 to 50 msec or so.
> For the second method, if we mix speech and syn tone at a
> given SNR we can recover
> speech by subtracting the syn tone at the received side,
> and the quality is not much less than the original I believe
Sorry but voice quality will be significantly reduced. If your method was possible then it would have already been
used a long time ago.
> Audio watermarking technique is new concept to me, it seems
> interesting and I have just taken a look, but it sounds
> just suitable for digital audio media like audio files,...
> for transmitting signal like telephone line, I guess it
> will have some problems like tramission quality, noise and
> for spread spectrum method, how to get the reference
> (original speech) to get back watermarking data from the
> processed signal.
Yes I agree that it would be difficult to implement watermarking reliably for real-time telephony communications. But
the main idea -- "hiding" data inside the main audio signal, via spread spectrum, low bitrate communication -- might
be something that Nadav can use. He doesn't need to embed much, just a "1 or 0" marker, which he could toggle every
frame, and then exclusive-OR successive frames until he finds the precise location in each frame that consistently
toggles.
-Jeff
> --- On Thu, 12/3/09, Jeff Brower wrote:
>
> From: Jeff Brower
> Subject: Re: [audiodsp] Audio signal sync
> To: "Duc Nguyen Anh"
> Cc: a...
> Date: Thursday, December 3, 2009, 5:52 AM
>
>
>
> Duc-
>
>> First thank Jeff Brower very much for your clear explanation
>
>> to my answer why we have to use windows for signal
>
>> processing before.
>
>>> For Nadav Haklai's question, i think it makes sense in some
>
>> cases, for example when we want to encrypt the telephone
>
>> voice, do not want other people can hear it, to my experince
>
>> the way to do that is quite the same as Nadav expressed,
>
>> we do some changes, of course the result is still in the
>
>> voice band, and at the receiver side we undo the change to
>
>> recover the original speech.
>
>>> Base on the way Jeff gave, I believe we can do that without
>
>> affecting the voice quality. We still use tone in voice
>
>> band, for telephone often 0-4K. There are 2 ways:
>
>> 1) we send first only the sync tone for curtain interval of
>
>> time and after that only the voice. At the receiver side,
>
>> we base on the received signal, used some kind of filers,
>
>> it is easy to know whether the coming signal is sync tone or
>
>> voice(based spectral for example), if it is syn tone, do
>
>> recover tone by VCO for example and get time-synchronized
>
>> with the sender.
>
>> 2) We send both syn tone and voice but with known SNR of syn
>
>> tone and voice, assume the transmission line is good
>
>> enough, we can recover the original speech and extract the
>
>> tone by spectral subtraction (the simplest) and do the same
>
>> as the 1)
>
>>> Above is what I think, if there is any thing wrong please
>
>> feel free to tell me
>
> These methods are not a good idea to implement for the same reason that telephone companies never combine signaling
>
> and voice at the same time: voice quality is unacceptable. Even going the other way -- adding voice to signaling --
>
> is a notoriously difficult problem (you can search on "talk off" to study this).
>
> To convey some amount of non-speech data in-band, you might look at audio watermarking. Watermarking techniques
>
> "spectrally spread" small amounts of data (such as coypright info), and mix with audio. The idea is the data is added
>
> as very low level white noise and is (hopefully) imperceptible to listeners. But I don't think the time precision in
>
> this method would be good enough for the exact frame synchronization that Nadav needs. Also watermarking would
>
> significantly increase the DSP resources required (MIPS and memory).
>
> -Jeff
>
>> --- On Tue, 12/1/09, Jeff Brower wrote:
>
>>> From: Jeff Brower > Subject: Re: [audiodsp] Audio signal sync
>
>> To: "Nadav Haklai" > Cc: audiodsp@yahoogroup s.com
>
>> Date: Tuesday, December 1, 2009, 9:52 AM
>
>>>>>>>>> Â
>
>>>>>>>>>>> Nadav-
>
>>>>>> That is correct. For now, the project is intended for some sort of
>
>>>> demonstration.
>
>>>> So the telephone line noise is irrelevant (i'll be only using 1 meter std
>
>>>> audio line).
>
>>>> Although, i was hoping to find a way to sync those two systems without any
>
>>>> other link besides the audio line.
>
>>>> I guessed that syncing framed audio signals between systems is a common
>
>>>> problem and there must be a simple solution (using special sync frame which
>
>>>> can be played and detected, adding FSK to the unused bw etc)
>
>>>>> I don't want to sound negative, as you're trying hard and seem to be persistent. But we have to apply rational
>
>>> thinking, not wishful thinking. I would ask you this: if you think this is a common method, then what audio
>> systems
>
>>> can you name that do this?
>
>>>>> The basic problem you're facing: if you modify your audio signal "in band", then you will significantly affect the
>
>>> quality of what people hear. Whatever modification you would make needs substantial energy to be detected reliably
>
>>> and accurately enough to provide precise frame synchronization -- listeners *will* hear it. But you haven't
>> specified
>
>>> what quality you need and what degradation you're willing to accept.
>
>>>>> RF communication systems often require precise synchronization. RF methods typically use bandwidth outside of the
>
>>> audio or data signal. For example, if you sample at 16 kHz but only transmit telephony quality audio (i.e. speech)
>
>>> limited to 4 kHz bandwidth, then you might use the upper 4 kHz for "out of band" information. You could send a 7
>> kHz
>
>>> tone short burst every frame, detect that and use its center, or peak power, as the frame marker. In your
>
>>> receive-side DSP code you would apply a sharp LPF to remove this from the audio signal.
>
>>>>> -Jeff
>
>>>>>> On Tue, Dec 1, 2009 at 12:39 AM, Jaime Andres Aranguren Cardona <
>
>>>> jaime_aranguren@ yahoo.com> wrote:
>
>>>>>>>> Hi all,
>
>>>>>>>>> Although Jeff's point is a certain one, maybe the project is kind of demo,
>
>>>>> and needs some sort of link (say a DSP's serial port) to demonstrate that
>
>>>>> the encoding and decoding is working on the DSP sides, and the link is out
>
>>>>> of the question. The issue here is that the link is what is causing the
>
>>>>> trouble. In other words, the problem is to get to DSPs to talk to each other
>
>>>>> through some link, that link can be a serial port (of the kind where people
>
>>>>> usually attach ADCs, DACs, Codecs) or an SPI peripheral on the DSPs or TWI,
>
>>>>> UARTs, whatever. And if this is the case, usually the HRM (Hardware
>
>>>>> Reference Manual) is all what the OP (original poster) needs.
>
>>>>>>>>> Maybe I am 100% lost here, but this is how I interpreted the question.
>
>>>>>>>>> Regards,
>
>>>>>>>>>>>>> Jaime Andrés Aranguren Cardona
>
>>>>> jaime.aranguren@ ieee.org
>
>>>>> jaime.aranguren@ computer. org
>
>>>>>>>>>>>>> ------------ --------- ---------
>
>>>>> *Von:* Jeff Brower >>>> *An:* Nadav >>>> *CC:* audiodsp@yahoogroup s.com
>
>>>>> *Gesendet:* Montag, den 30. November 2009, 19:00:55 Uhr
>
>>>>> *Betreff:* Re: [audiodsp] Audio signal sync
>
>>>>>>>>>>>>>>>>> Nadav-
>
>>>>>>>>> > I'm developing a real-time audio processing project which
>
>>>>> > consists of 2 DSP separated systems.
>
>>>>> > One system is encoder and other decoder.
>
>>>>> > between the systems there is a regular audio line.
>
>>>>> > the encoder samples the input audio data (8/16/24 bits,
>
>>>>> > 16Khz sampling rate) separates it into frames and
>
>>>>> > process each frame individually. The processed data is
>
>>>>> > played by the D/A to the audio line and is the input to
>
>>>>> > the decoder system.
>
>>>>> > the decoder system samples the data and decode the data.
>
>>>>> >>>>> > My problem is that the data is being processed in separate
>
>>>>> > frames and the decoder must know when each frame begin in
>
>>>>> > order to process it correctly.
>
>>>>> > How can i sync those 2 system using only the audio signal
>
>>>>> > line which carries the actual data?
>
>>>>> > Is it possible some data bits to the data somehow?
>
>>>>> > Can i use special frames to sync?
>
>>>>>>>>> Something is not making sense here. Normally two systems doing audio
>
>>>>> encode/decode communicate via digital transport
>
>>>>> -- that's the whole point of audio compression, right? It should like this:
>
>>>>>>>>> _______ Dig I/O _______
>
>>>>> Audio <--> | codec | <-------> | codec | <--> Audio
>
>>>>> |_______| |_______|
>
>>>>>>>>> where digital I/O is network (e.g. VoIP), RF (e.g. GSM, WiFi, satellite,
>
>>>>> etc), serial port (e.g. RS-422), etc.
>
>>>>>>>>> If you're connecting two systems via audio, then why do you need
>
>>>>> compression?
> The application I wanted to express is like this, you and I
> want to communicate safely through the public telephone
> service, we dont want any other can eavesdrop what we
> exchange, and of course we dont want the use some special
> service from telephone providers, we just design 2 devices
> (phones) as the normal ones added some features help us to
> do that.
>
> About quality of voice, for the first way I said, syn tone
> (signaling) is just sent the starting time for
> synchronization purpose, after that only voice is sent and
> the tone is removed by device so it is ok for voice quality
> but it comes with a bit delay.
What purpose is served by a "one time" sync tone? Nadav wants to do frame-by-frame synchronization. That would mean
a sync tone every 10 to 50 msec or so.
> For the second method, if we mix speech and syn tone at a
> given SNR we can recover
> speech by subtracting the syn tone at the received side,
> and the quality is not much less than the original I believe
Sorry but voice quality will be significantly reduced. If your method was possible then it would have already been
used a long time ago.
> Audio watermarking technique is new concept to me, it seems
> interesting and I have just taken a look, but it sounds
> just suitable for digital audio media like audio files,...
> for transmitting signal like telephone line, I guess it
> will have some problems like tramission quality, noise and
> for spread spectrum method, how to get the reference
> (original speech) to get back watermarking data from the
> processed signal.
Yes I agree that it would be difficult to implement watermarking reliably for real-time telephony communications. But
the main idea -- "hiding" data inside the main audio signal, via spread spectrum, low bitrate communication -- might
be something that Nadav can use. He doesn't need to embed much, just a "1 or 0" marker, which he could toggle every
frame, and then exclusive-OR successive frames until he finds the precise location in each frame that consistently
toggles.
-Jeff
> --- On Thu, 12/3/09, Jeff Brower wrote:
>
> From: Jeff Brower
> Subject: Re: [audiodsp] Audio signal sync
> To: "Duc Nguyen Anh"
> Cc: a...
> Date: Thursday, December 3, 2009, 5:52 AM
>
>
>
> Duc-
>
>> First thank Jeff Brower very much for your clear explanation
>
>> to my answer why we have to use windows for signal
>
>> processing before.
>
>>> For Nadav Haklai's question, i think it makes sense in some
>
>> cases, for example when we want to encrypt the telephone
>
>> voice, do not want other people can hear it, to my experince
>
>> the way to do that is quite the same as Nadav expressed,
>
>> we do some changes, of course the result is still in the
>
>> voice band, and at the receiver side we undo the change to
>
>> recover the original speech.
>
>>> Base on the way Jeff gave, I believe we can do that without
>
>> affecting the voice quality. We still use tone in voice
>
>> band, for telephone often 0-4K. There are 2 ways:
>
>> 1) we send first only the sync tone for curtain interval of
>
>> time and after that only the voice. At the receiver side,
>
>> we base on the received signal, used some kind of filers,
>
>> it is easy to know whether the coming signal is sync tone or
>
>> voice(based spectral for example), if it is syn tone, do
>
>> recover tone by VCO for example and get time-synchronized
>
>> with the sender.
>
>> 2) We send both syn tone and voice but with known SNR of syn
>
>> tone and voice, assume the transmission line is good
>
>> enough, we can recover the original speech and extract the
>
>> tone by spectral subtraction (the simplest) and do the same
>
>> as the 1)
>
>>> Above is what I think, if there is any thing wrong please
>
>> feel free to tell me
>
> These methods are not a good idea to implement for the same reason that telephone companies never combine signaling
>
> and voice at the same time: voice quality is unacceptable. Even going the other way -- adding voice to signaling --
>
> is a notoriously difficult problem (you can search on "talk off" to study this).
>
> To convey some amount of non-speech data in-band, you might look at audio watermarking. Watermarking techniques
>
> "spectrally spread" small amounts of data (such as coypright info), and mix with audio. The idea is the data is added
>
> as very low level white noise and is (hopefully) imperceptible to listeners. But I don't think the time precision in
>
> this method would be good enough for the exact frame synchronization that Nadav needs. Also watermarking would
>
> significantly increase the DSP resources required (MIPS and memory).
>
> -Jeff
>
>> --- On Tue, 12/1/09, Jeff Brower wrote:
>
>>> From: Jeff Brower > Subject: Re: [audiodsp] Audio signal sync
>
>> To: "Nadav Haklai" > Cc: audiodsp@yahoogroup s.com
>
>> Date: Tuesday, December 1, 2009, 9:52 AM
>
>>>>>>>>> Â
>
>>>>>>>>>>> Nadav-
>
>>>>>> That is correct. For now, the project is intended for some sort of
>
>>>> demonstration.
>
>>>> So the telephone line noise is irrelevant (i'll be only using 1 meter std
>
>>>> audio line).
>
>>>> Although, i was hoping to find a way to sync those two systems without any
>
>>>> other link besides the audio line.
>
>>>> I guessed that syncing framed audio signals between systems is a common
>
>>>> problem and there must be a simple solution (using special sync frame which
>
>>>> can be played and detected, adding FSK to the unused bw etc)
>
>>>>> I don't want to sound negative, as you're trying hard and seem to be persistent. But we have to apply rational
>
>>> thinking, not wishful thinking. I would ask you this: if you think this is a common method, then what audio
>> systems
>
>>> can you name that do this?
>
>>>>> The basic problem you're facing: if you modify your audio signal "in band", then you will significantly affect the
>
>>> quality of what people hear. Whatever modification you would make needs substantial energy to be detected reliably
>
>>> and accurately enough to provide precise frame synchronization -- listeners *will* hear it. But you haven't
>> specified
>
>>> what quality you need and what degradation you're willing to accept.
>
>>>>> RF communication systems often require precise synchronization. RF methods typically use bandwidth outside of the
>
>>> audio or data signal. For example, if you sample at 16 kHz but only transmit telephony quality audio (i.e. speech)
>
>>> limited to 4 kHz bandwidth, then you might use the upper 4 kHz for "out of band" information. You could send a 7
>> kHz
>
>>> tone short burst every frame, detect that and use its center, or peak power, as the frame marker. In your
>
>>> receive-side DSP code you would apply a sharp LPF to remove this from the audio signal.
>
>>>>> -Jeff
>
>>>>>> On Tue, Dec 1, 2009 at 12:39 AM, Jaime Andres Aranguren Cardona <
>
>>>> jaime_aranguren@ yahoo.com> wrote:
>
>>>>>>>> Hi all,
>
>>>>>>>>> Although Jeff's point is a certain one, maybe the project is kind of demo,
>
>>>>> and needs some sort of link (say a DSP's serial port) to demonstrate that
>
>>>>> the encoding and decoding is working on the DSP sides, and the link is out
>
>>>>> of the question. The issue here is that the link is what is causing the
>
>>>>> trouble. In other words, the problem is to get to DSPs to talk to each other
>
>>>>> through some link, that link can be a serial port (of the kind where people
>
>>>>> usually attach ADCs, DACs, Codecs) or an SPI peripheral on the DSPs or TWI,
>
>>>>> UARTs, whatever. And if this is the case, usually the HRM (Hardware
>
>>>>> Reference Manual) is all what the OP (original poster) needs.
>
>>>>>>>>> Maybe I am 100% lost here, but this is how I interpreted the question.
>
>>>>>>>>> Regards,
>
>>>>>>>>>>>>> Jaime Andrés Aranguren Cardona
>
>>>>> jaime.aranguren@ ieee.org
>
>>>>> jaime.aranguren@ computer. org
>
>>>>>>>>>>>>> ------------ --------- ---------
>
>>>>> *Von:* Jeff Brower >>>> *An:* Nadav >>>> *CC:* audiodsp@yahoogroup s.com
>
>>>>> *Gesendet:* Montag, den 30. November 2009, 19:00:55 Uhr
>
>>>>> *Betreff:* Re: [audiodsp] Audio signal sync
>
>>>>>>>>>>>>>>>>> Nadav-
>
>>>>>>>>> > I'm developing a real-time audio processing project which
>
>>>>> > consists of 2 DSP separated systems.
>
>>>>> > One system is encoder and other decoder.
>
>>>>> > between the systems there is a regular audio line.
>
>>>>> > the encoder samples the input audio data (8/16/24 bits,
>
>>>>> > 16Khz sampling rate) separates it into frames and
>
>>>>> > process each frame individually. The processed data is
>
>>>>> > played by the D/A to the audio line and is the input to
>
>>>>> > the decoder system.
>
>>>>> > the decoder system samples the data and decode the data.
>
>>>>> >>>>> > My problem is that the data is being processed in separate
>
>>>>> > frames and the decoder must know when each frame begin in
>
>>>>> > order to process it correctly.
>
>>>>> > How can i sync those 2 system using only the audio signal
>
>>>>> > line which carries the actual data?
>
>>>>> > Is it possible some data bits to the data somehow?
>
>>>>> > Can i use special frames to sync?
>
>>>>>>>>> Something is not making sense here. Normally two systems doing audio
>
>>>>> encode/decode communicate via digital transport
>
>>>>> -- that's the whole point of audio compression, right? It should like this:
>
>>>>>>>>> _______ Dig I/O _______
>
>>>>> Audio <--> | codec | <-------> | codec | <--> Audio
>
>>>>> |_______| |_______|
>
>>>>>>>>> where digital I/O is network (e.g. VoIP), RF (e.g. GSM, WiFi, satellite,
>
>>>>> etc), serial port (e.g. RS-422), etc.
>
>>>>>>>>> If you're connecting two systems via audio, then why do you need
>
>>>>> compression?
Reply by ●December 4, 20092009-12-04
Hi Jeff,
Just try to find a way that can handle the problems, to my experience, this has applications in our real life. After rethinking about what I suggested I realize that the second case is not feasible to implement due to the extraction of the syn tone and get synchronized with it. The first case is just one-time synchronization so we can not obtain frame to frame synchronization. However, we can use some tricks, for example, we process the voice to limit it just in range of 300 Hz to 3400 Hz (quality still good), choose the syn tone in range 0-300 Hz or 3400 - 4000 Hz (keep away from 50,60 Hz and their harmonics due to the characteristics of electricity system) and add it to the voice. One problem with this approach is that since we can not recover exactly the syn tone at receiver side, so we have to accept some error (deviation), and should pay attention to this when do processing.
Looking forward to hearing your comments
Have a nice weekend to all of you
Just try to find a way that can handle the problems, to my experience, this has applications in our real life. After rethinking about what I suggested I realize that the second case is not feasible to implement due to the extraction of the syn tone and get synchronized with it. The first case is just one-time synchronization so we can not obtain frame to frame synchronization. However, we can use some tricks, for example, we process the voice to limit it just in range of 300 Hz to 3400 Hz (quality still good), choose the syn tone in range 0-300 Hz or 3400 - 4000 Hz (keep away from 50,60 Hz and their harmonics due to the characteristics of electricity system) and add it to the voice. One problem with this approach is that since we can not recover exactly the syn tone at receiver side, so we have to accept some error (deviation), and should pay attention to this when do processing.
Looking forward to hearing your comments
Have a nice weekend to all of you
