Open Holzhaus opened 10 years ago
That's pretty convincing... I'd definitely be in favour of an STT-plugin, at the very least. Really curious to see how it performs.
+1, the pi 2 has 900MHz and 1Gb Ram, everything is running fast for me. I'm curious to see how it performs
Looks like Sphinx4 still not possible to run on RPi2B. I tried Sphinx4-5prealpha and it need too much memory...
I wrote a simple http web server to host spinx4-5prealpha on a network m/c, Jasper on RPi2B successfully connect to it and use it's STT (just like Google STT) service. But one thing I mark, Speech-To-Text is not that effiecint using Sphinx4 with default Acoustic Model, LM and Dict. To improve efficincy we need to provide custom LM and Dict.
Jasper Debug Log:-
INFO:requests.packages.urllib3.connectionpool:Starting new HTTP connection (1): Sphinx4WebServer 200 OK time now INFO:client.stt:Sphinx4 Transcribed: u'TIME NOW' DEBUG:client.brain:'TIME' is a valid phrase for module 'Time' DEBUG:client.tts:Saying 'It is 4:40 PM right now.' with 'pico-tts' DEBUG:client.tts:Executing pico2wave --wave /tmp/tmpdYRKnn.wav -l en-GB 'It is 4:40 PM right now.' DEBUG:client.tts:Executing aplay /tmp/tmpdYRKnn.wav DEBUG:client.tts:Output was: 'Playing WAVE '/tmp/tmpdYRKnn.wav' : Signed 16 bit Little Endian, Rate 16000 Hz, Mono
I just saw a comparison of pocketsphinx and sphinx4. It look's like the recognition of sphinx4 is lightyears ahead of pocketsphinx'.
I hate java, especially on low-end hardware like the RPi, but maybe a Sphinx4-STT-plugin is possible nevertheless.
To archieve this, I think some of this code could be recycled.
What do you think?