DSPRelated.com
Forums

melp synthesis silence output

Started by steh...@hotmail.com July 11, 2007
Hi,
I am using a codec with sampling frequency of 32KHz, I am decimating the input to get 8KHz frequency. These samples are then fed into the MELP Analysis. The ouput of melp analysis is transmitted through a modem to the remote side. Here MELP Synthesis is called, and the speech is played by codec. I have 2 questions here,

1. As the data is being transmitted to remote side, there can be some corruption. Is it possible that melp synthesis variables gets so badly corrupted, that after synthesizing these corrupted frames, it cannot synthsize correct frames too???

2. If I give a sinosidal waveform, for the sake of testing my codec amplitude etc, what output should I expect. One thing I am sure about is that it wont be sinosidal, as melp is designed for speech, but still I would like to know what should be its output?

Thanks and Regards,
Saba
Saba-

> I am using a codec with sampling frequency of 32KHz, I am decimating the
> input to get 8KHz frequency. These samples are then fed into the MELP
> Analysis. The ouput of melp analysis is transmitted through a modem to
> the remote side. Here MELP Synthesis is called, and the speech is played
> by codec. I have 2 questions here,
>
> 1. As the data is being transmitted to remote side, there can be some
> corruption. Is it possible that melp synthesis variables gets so badly
> corrupted, that after synthesizing these corrupted frames, it cannot
> synthsize correct frames too???

If transmission errors affect the sync bit, then you might not get valid synthesis
results until after several consecutive frames go through with no errors. It can
take several hundred msec for the MELPe algorithm to re-acquire sync at 2400 bps.

> 2. If I give a sinosidal waveform, for the sake of testing my codec
> amplitude etc, what output should I expect. One thing I am sure about
> is that it wont be sinosidal, as melp is designed for speech, but still
> I would like to know what should be its output?

A sine waveform will sound garbled, not like anything I can describe. Just will
sound bad.

-Jeff
Thanks for your reply.
The problem that I am facing is that after processing some corrupted frames MELP sysnthesis starts giving complete silence at the output. The thing that puzzles me is that it is not recovering from the errors, as I am sending one saved packet of DTMF tone. I have observed that even when melp starts receving the correct from it does not recover. As I am using the same frame at each interval, is it possible that it is not finding the sync bit right and never re syncs again.

>From the output of the sinosidal wave, I mean that if I want to see the output waveform on an oscilloscope after passing the digital data through MELP and DAC, what kind of wave form should I expect???

Thanks and Regards,

Saba
Saba-

> Thanks for your reply.
> The problem that I am facing is that after processing some corrupted frames MELP sysnthesis starts giving complete
> silence at the output. The thing that puzzles me is that it is not recovering from the errors, as I am sending one
> saved packet of DTMF tone. I have observed that even when melp starts receving the correct from it does not recover.
> As I am using the same frame at each interval, is it possible that it is not finding the sync bit right and never re
> syncs again.

I don't know. It should recover, as long as your frames are correctly formatted, including sync bit. The classic
test is to use a serial bitstream transmission method (e.g. RS-422), pull the connector (upon which MELPe synthesis
output becomes garbled, "helicopter noise", etc), then reconnect. After a half second to several seconds, it should
be back in sync and reconstructed results sound good. It shouldn't matter what junk was received in the interim.

As far as losing sync, how are you sending frames? Is it a serial bitstream? You can't lose sync if you're always
sending frames on byte boundaries, for example bus type of interface. It has to be bit-by-bit.

> From the output of the sinosidal wave, I mean that if I want to see the output waveform on an oscilloscope after
> passing the digital data through MELP and DAC, what kind of wave form should I expect???

Nothing that can be defined. Just junk.

-Jeff