update

嘉渊

2023-04-24 6427c834dfd97b1f05c6659cdc7ccf010bf82fe1

 egs_modelscope/asr/paraformer/speech_paraformer_asr_nat-zh-cn-16k-common-vocab8404-online/infer.py

@@ -1,52 +1,37 @@
import os
import logging
import torch
import torchaudio
import soundfile

from modelscope.pipelines import pipeline
from modelscope.utils.constant import Tasks
from modelscope.utils.logger import get_logger

logger = get_logger(log_level=logging.CRITICAL)
logger.setLevel(logging.CRITICAL)

os.environ["MODELSCOPE_CACHE"] = "./"
inference_pipeline = pipeline(
    task=Tasks.auto_speech_recognition,
    model='damo/speech_paraformer_asr_nat-zh-cn-16k-common-vocab8404-online',
    model_revision='v1.0.2')

waveform, sample_rate = torchaudio.load("asr_example_zh.wav")
speech_length = waveform.shape[1]
speech = waveform[0]
model_dir = os.path.join(os.environ["MODELSCOPE_CACHE"], "damo/speech_paraformer_asr_nat-zh-cn-16k-common-vocab8404-online")
speech, sample_rate = soundfile.read(os.path.join(model_dir, "example/asr_example.wav"))
speech_length = speech.shape[0]

cache_en = {"start_idx": 0, "pad_left": 0, "stride": 10, "pad_right": 5, "cif_hidden": None, "cif_alphas": None}
cache_de = {"decode_fsmn": None}
cache = {"encoder": cache_en, "decoder": cache_de}
param_dict = {}
param_dict["cache"] = cache

first_chunk = True
speech_buffer = speech
speech_cache = []
sample_offset = 0
step = 4800  #300ms
param_dict = {"cache": dict(), "is_final": False}
final_result = ""

while len(speech_buffer) > 0:
    if first_chunk:
        if len(speech_buffer) >= 14400:
            rec_result = inference_pipeline(audio_in=speech_buffer[0:14400], param_dict=param_dict)
            speech_buffer = speech_buffer[4800:]
        else:
            cache_en["stride"] = len(speech_buffer) // 960
            cache_en["pad_right"] = 0
            rec_result = inference_pipeline(audio_in=speech_buffer, param_dict=param_dict)
            speech_buffer = []
        cache_en["start_idx"] = -5
        first_chunk = False
    else:
        cache_en["start_idx"] += 10
        if len(speech_buffer) >= 4800:
            cache_en["pad_left"] = 5
            rec_result = inference_pipeline(audio_in=speech_buffer[:19200], param_dict=param_dict)
            speech_buffer = speech_buffer[9600:]
        else:
            cache_en["stride"] = len(speech_buffer) // 960
            cache_en["pad_right"] = 0
            rec_result = inference_pipeline(audio_in=speech_buffer, param_dict=param_dict)
            speech_buffer = []
    if rec_result['text'] != "sil":
for sample_offset in range(0, speech_length, min(step, speech_length - sample_offset)):
    if sample_offset + step >= speech_length - 1:
        step = speech_length - sample_offset
        param_dict["is_final"] = True
    rec_result = inference_pipeline(audio_in=speech[sample_offset: sample_offset + step],
                                    param_dict=param_dict)
    if len(rec_result) != 0 and rec_result['text'] != "sil" and rec_result['text'] != "waiting_for_more_voice":
        final_result += rec_result['text']
    print(rec_result)
print(final_result)