deepspeech_bot.py

import os
import logging
import subprocess
import sys

import m3
import m3.lib.ws
from m3.lib.model import media_table
from m3.natlang.blockly import register_custom_toolbox
from pgaccess import pgsql


logger = logging.getLogger('deepspeech')
pool = m3.lib.ws.Pool('deepspeech')


@pgsql.transaction
def s2t(media_id):
    devnull = open(os.devnull, 'w')
    wav_name = os.path.expanduser('~/audio/%s.wav' % media_id)
    data = media_table.select('content').where(id=media_id).scalar()
    with open(wav_name, 'wb') as fout:
        fout.write(data)
    out = subprocess.Popen('/home/m3/deepspeech.env/bin/deepspeech '
                           '--model ~/italian/output_graph.pbmm '
                           '--lm ~/italian/lm.binary '
                           '--trie ~/italian/trie '
                           '--alphabet ~/italian/alphabet.txt '
                           '--audio %s' % wav_name,
                           stdout=subprocess.PIPE,
                           stderr=devnull,
                           shell=True).communicate()[0]
    return out


@pool.delegate
def delegated_s2t(media_id):
    return s2t(media_id)


def deepspeech():
    last_msg = m3.session.received_message
    if last_msg is None:
        logger.error('last message not present')
        return
    media_id = last_msg.metadata.get('audio')
    if media_id is None:
        logger.error('last message has no audio')
        return
    txt = delegated_s2t(media_id)
    return dict(txt=txt, media_id=media_id)


def main():
    print s2t(sys.argv[1])


def init_m3():
    register_custom_toolbox('deepspeech', deepspeech)


if __name__ == '__main__':
    main()