Hi,

On 28/02/18 16:53, Ariane stolfi wrote:

I'm planning to work on a project that will use a microphone to query on freesound database, and we are planning to use speech recognition to send the queries.  We will use pure data installed in a Bella board and e-textile sensors to process the sound.

I'm considering send audio and do the speech recognition in a server using the Google API or use Sphynx to do the recognition in Pure data. Does anyone here has any experience with this recently and can make a suggestion?

I used pocket_sphinx from the command line in one project (that didn't use Pd)[1] The key line for speech transcription in my shell script is: utterance="$(pocketsphinx_continuous -infile utterance.wav 2>/dev/null)" The input was synthetic sound, so rather noise-free ideal conditions.

I imagine that you could use Pd to do something like detect presence of sound and save the segment to a wav file, calling out to Sphinx with a scripting external. The same process would be necessary for Google API I suppose, unless you intend to be streaming everything there.

Claude

[1] https://mathr.co.uk/blog/2017-04-08_divergent_protocol.html

https://mathr.co.uk