[Speex-dev] [Fwd: Re: Which product to use?]
David M. Witten II
wittend at wwrinc.com
Tue Jun 23 16:44:39 PDT 2009
I have been interested in some of the same questions that you have been
asking. A while back I asked a person at VLSI ( http://www.vlsi.fi/ )
in Finland if they had a product that might work. The exchange of email
is included below.
In summary, the answer was probably no, but maybe in the future. The
VS1005 chip he mentions does not seem to have appeared yet.
I currently have a pair of Microchip DSPIC33 based boards with their
optimized/proprietized implementation of Speex on hand for testing, but
I am looking for alternate solutions. I don't believe that Microchip's
code is going to allow me to experiment with the 2.15k bit rate mode
that is most attractive for my purposes. I understand that that mode
has serious limitations and needs a variety of modifications, at least
in the applications that interest me.
-- Dave Witten
> -------- Original Message --------
> Thanks Pekka,
> Your response is very helpful. I will look closely at the VS1053 and
> watch for the VS1005 to see what features it offers. I have an
> implementation of Speex for one family of integer 16 bit DSP's, with
> information on code size for that environment. There are some problems
> with Speex for my specific application, but so far it seems as good a
> place as any to begin.
> The trick is to keep the project free of IP restrictions as there are
> potential problems with using restricted algorithms in certain amateur
> radio contexts (even beyond my own bias towards open source). Some
> regulations in the US and some of the rules relating to amateur
> satellite communications can be read to preclude the use of existing
> proprietary algorithms.
> At present the low-bit rate algorithms known to be best suited to
> amateur communications in the HF and VHF regions are AMBE and MELP, both
> of which are seriously encumbered. An open source solution suited to
> the peculiarities of this market would be welcomed and it is easy to
> demonstrate that a chip based implementation would find a place in a
> number of amateur digital voice devices.
> I will probably order a proto board in the near future to see what it
> has to offer.
> -- Dave Witten
> Audio Solution Support wrote:
>> Dear David,
>> Sorry for the delay.
>> We have done some work on Speex and it is possible that the upcoming
>> VS1005 will support it. I think that VS1053 might have enough memory to
>> do this. However I think that it will not be a trivial task. Notice that
>> while VS1000 might seem suitable it does not have an AD-converter.
>> It is true that the Speex should not be as complex as some other
>> decoders we have. Usually however the decoders are in ROM. Our devices
>> have much more ROM than they have RAM. Compared to for example FLASH,
>> ROM is cheap and makes it possible to have a lot of features in an
>> affordable device.
>> One possibility would be to use low-bitrate OggVorbis. It is not the
>> optimal for speech encoding but give rather good results with proper
>> implementation. We have an OggVorbis encoder application for VS1053.
>> Take a look at: http://www.vlsi.fi/en/products/vs1053.html and
>> If you plan to use your own host uC then suitable test board would be
>> the Prototyping Board
>> Best regards,
>> Audio Solution Support
>> David Witten wrote:
>>> I am designing a product for use by amateur radio operators to provide
>>> voice over narrow band RF.
>>> As conceived, the device would use a Speex codec implemented in
>>> firmware to encode speech to be modulated onto the RF carrier for
>>> transmission, and decode the data stream demodulated from the
>>> receiver's output for processing back into audible form. Nothing
>>> revolutionary, but I would like to put it all into a compact package
>>> with simple setup aimed at the needs of this particular market.
>>> My question is: which, if any, of your chip level and/or demo board
>>> products might help me in this project?
>>> My understanding is that Speex is similar to but less resource
>>> intensive than many of the algorithms that you currently supply in
>>> your products. Is it realistic to think that it could be implemented
>>> in something like the VS1000? Are you aware of anyone who has done
>>> this? Can your VOIP speaker phone evaluation board be modified to
>>> demonstrate this functionality?
>>> Or am I just clueless and asking the wrong questions?...
More information about the Speex-dev