vosk_asr_server.py

# implementation of VOSK server voice recognition
# origin: https://github.com/alphacep/vosk-server/blob/master/websocket/asr_server.py
# license: Apache 2.0

# Запускать после первого запуска WEBAPI!! - т.к. использует для настроек файл runva_webapi.json
# для размещения использует порт webapi+1 и тот же хост

import json
import os
import sys
import asyncio
import pathlib
import websockets
import concurrent.futures
import logging
from vosk import Model, SpkModel, KaldiRecognizer

version = "1.1"

# main options
with open('runva_webapi.json', 'r', encoding="utf-8") as f:
    s = f.read(10000000)
    f.close()
webapi_options = json.loads(s)

def process_chunk(rec, message):
    # with open('temp/asr_server_test.wav', 'wb') as the_file:
    #     the_file.write(message)

    if message == '{"eof" : 1}':
        return rec.FinalResult(), True
    elif rec.AcceptWaveform(message):
        res = rec.Result()
        #print("Result:",res)
        return res, False
    else:
        res = rec.PartialResult()
        #print("Part Result:",res)
        return rec.PartialResult(), False

async def recognize(websocket, path):
    global model
    global spk_model
    global args
    global pool

    loop = asyncio.get_running_loop()
    rec = None
    phrase_list = None
    sample_rate = args.sample_rate
    show_words = args.show_words
    max_alternatives = args.max_alternatives

    logging.info('Connection from %s', websocket.remote_address)

    while True:

        message = await websocket.recv()

        # Load configuration if provided
        if isinstance(message, str) and 'config' in message:
            jobj = json.loads(message)['config']
            logging.info("Config %s", jobj)
            if 'phrase_list' in jobj:
                phrase_list = jobj['phrase_list']
            if 'sample_rate' in jobj:
                sample_rate = float(jobj['sample_rate'])
            if 'words' in jobj:
                show_words = bool(jobj['words'])
            if 'max_alternatives' in jobj:
                max_alternatives = int(jobj['max_alternatives'])
            continue

        # Create the recognizer, word list is temporary disabled since not every model supports it
        if not rec:
            if phrase_list:
                rec = KaldiRecognizer(model, sample_rate, json.dumps(phrase_list, ensure_ascii=False))
            else:
                rec = KaldiRecognizer(model, sample_rate)
            rec.SetWords(show_words)
            rec.SetMaxAlternatives(max_alternatives)
            if spk_model:
                rec.SetSpkModel(spk_model)

        response, stop = await loop.run_in_executor(pool, process_chunk, rec, message)
        await websocket.send(response)
        if stop: break



async def start():

    global model
    global spk_model
    global args
    global pool

    # Enable loging if needed
    #
    # logger = logging.getLogger('websockets')
    # logger.setLevel(logging.INFO)
    # logger.addHandler(logging.StreamHandler())
    logging.basicConfig(level=logging.INFO)

    args = type('', (), {})()

    # для размещения использует порт webapi+1 и тот же хост
    args.interface = os.environ.get('VOSK_SERVER_INTERFACE', webapi_options["host"])
    args.port = int(os.environ.get('VOSK_SERVER_PORT', int(webapi_options["port"])+1))
    args.model_path = os.environ.get('VOSK_MODEL_PATH', 'model')
    args.spk_model_path = os.environ.get('VOSK_SPK_MODEL_PATH')
    args.sample_rate = float(os.environ.get('VOSK_SAMPLE_RATE', 16000))
    args.max_alternatives = int(os.environ.get('VOSK_ALTERNATIVES', 0))
    args.show_words = bool(os.environ.get('VOSK_SHOW_WORDS', True))

    if len(sys.argv) > 1:
        args.model_path = sys.argv[1]

    # Gpu part, uncomment if vosk-api has gpu support
    #
    # from vosk import GpuInit, GpuInstantiate
    # GpuInit()
    # def thread_init():
    #     GpuInstantiate()
    # pool = concurrent.futures.ThreadPoolExecutor(initializer=thread_init)

    model = Model(args.model_path)
    spk_model = SpkModel(args.spk_model_path) if args.spk_model_path else None

    pool = concurrent.futures.ThreadPoolExecutor((os.cpu_count() or 1))

    print("VOSK Speech Recognition server v{0} started! host={1}, port={2}".format(version,args.interface,args.port))

    async with websockets.serve(recognize, args.interface, args.port):
        await asyncio.Future()


if __name__ == '__main__':
    asyncio.run(start())