蜜桃av色欲a片精品一区,麻豆aⅴ精品无码一区二区,亚洲人成网站在线播放影院在线,亚洲 素人 字幕 在线 最新

  新一代 Kaldi 實(shí)時(shí)語音識別： sherpa-ncnn Python API

發(fā)布日期：2025/5/27 15:52:02      瀏覽量：

新一代 Kaldi 實(shí)時(shí)語音識別： sherpa-ncnn Python API

GitHub - k2-fsa/sherpa-ncnn: Real-time speech recognition and voice activity detection (VAD) using next-gen Kaldi with ncnn without Internet connection. Support iOS, Android, Linux, macOS, Windows, Raspberry Pi, VisionFive2, LicheePi4A etc.

Py代碼：

#!/usr/bin/env python3

# Real-time speech recognition from a microphone with sherpa-ncnn Python API
#
# Please refer to
# https://k2-fsa.github.io/sherpa/ncnn/pretrained_models/index.html
# to download pre-trained models

import sys

try:
    import sounddevice as sd
except ImportError as e:
    print("Please install sounddevice first. You can use")
    print()
    print("  pip install sounddevice")
    print()
    print("to install it")
    sys.exit(-1)

import sherpa_ncnn


def create_recognizer():
    # Please replace the model files if needed.
    # See https://k2-fsa.github.io/sherpa/ncnn/pretrained_models/index.html
    # for download links.
    recognizer = sherpa_ncnn.Recognizer(
        tokens="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/tokens.txt",
        encoder_param="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/encoder_jit_trace-pnnx.ncnn.param",
        encoder_bin="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/encoder_jit_trace-pnnx.ncnn.bin",
        decoder_param="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/decoder_jit_trace-pnnx.ncnn.param",
        decoder_bin="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/decoder_jit_trace-pnnx.ncnn.bin",
        joiner_param="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/joiner_jit_trace-pnnx.ncnn.param",
        joiner_bin="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/joiner_jit_trace-pnnx.ncnn.bin",
        num_threads=4,
        hotwords_file="",
        hotwords_score=1.5,
    )
    return recognizer


def main():
    print("Started! Please speak")
    recognizer = create_recognizer()
    sample_rate = recognizer.sample_rate
    samples_per_read = int(0.1 * sample_rate)  # 0.1 second = 100 ms
    last_result = ""
    with sd.InputStream(channels=1, dtype="float32", samplerate=sample_rate) as s:
        while True:
            samples, _ = s.read(samples_per_read)  # a blocking read
            samples = samples.reshape(-1)
            recognizer.accept_waveform(sample_rate, samples)
            result = recognizer.text
            if last_result != result:
                last_result = result
                print("\r{}".format(result), end="", flush=True)


if __name__ == "__main__":
    devices = sd.query_devices()
    print(devices)
    default_input_device_idx = sd.default.device[0]
    print(f’Use default device: {devices[default_input_device_idx]["name"]}’)

    try:
        main()
    except KeyboardInterrupt:
        print("\nCaught Ctrl + C. Exiting")