enginex_bi_series-sherpa-onnx/python-api-examples/speech-recognition-from-microphone.py

#!/usr/bin/env python3

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

import argparse
import sys
from pathlib import Path

from typing import List

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

import sherpa_onnx


def assert_file_exists(filename: str):
    assert Path(filename).is_file(), (
        f"{filename} does not exist!\n"
        "Please refer to "
        "https://k2-fsa.github.io/sherpa/onnx/pretrained_models/index.html to download it"
    )


def get_args():
    parser = argparse.ArgumentParser(
        formatter_class=argparse.ArgumentDefaultsHelpFormatter
    )

    parser.add_argument(
        "--tokens",
        type=str,
        required=True,
        help="Path to tokens.txt",
    )

    parser.add_argument(
        "--encoder",
        type=str,
        required=True,
        help="Path to the encoder model",
    )

    parser.add_argument(
        "--decoder",
        type=str,
        required=True,
        help="Path to the decoder model",
    )

    parser.add_argument(
        "--joiner",
        type=str,
        help="Path to the joiner model",
    )

    parser.add_argument(
        "--decoding-method",
        type=str,
        default="greedy_search",
        help="Valid values are greedy_search and modified_beam_search",
    )

    parser.add_argument(
        "--max-active-paths",
        type=int,
        default=4,
        help="""Used only when --decoding-method is modified_beam_search.
        It specifies number of active paths to keep during decoding.
        """,
    )

    parser.add_argument(
        "--provider",
        type=str,
        default="cpu",
        help="Valid values: cpu, cuda, coreml",
    )

    parser.add_argument(
        "--hotwords-file",
        type=str,
        default="",
        help="""
        The file containing hotwords, one words/phrases per line, and for each
        phrase the bpe/cjkchar are separated by a space. For example:

        ▁HE LL O ▁WORLD
        你 好 世 界
        """,
    )

    parser.add_argument(
        "--hotwords-score",
        type=float,
        default=1.5,
        help="""
        The hotword score of each token for biasing word/phrase. Used only if
        --hotwords-file is given.
        """,
    )

    parser.add_argument(
        "--blank-penalty",
        type=float,
        default=0.0,
        help="""
        The penalty applied on blank symbol during decoding.
        Note: It is a positive value that would be applied to logits like
        this `logits[:, 0] -= blank_penalty` (suppose logits.shape is
        [batch_size, vocab] and blank id is 0).
        """,
    )

    parser.add_argument(
        "--hr-dict-dir",
        type=str,
        default="",
        help="If not empty, it is the jieba dict directory for homophone replacer",
    )

    parser.add_argument(
        "--hr-lexicon",
        type=str,
        default="",
        help="If not empty, it is the lexicon.txt for homophone replacer",
    )

    parser.add_argument(
        "--hr-rule-fsts",
        type=str,
        default="",
        help="If not empty, it is the replace.fst for homophone replacer",
    )

    return parser.parse_args()


def create_recognizer(args):
    assert_file_exists(args.encoder)
    assert_file_exists(args.decoder)
    assert_file_exists(args.joiner)
    assert_file_exists(args.tokens)
    # Please replace the model files if needed.
    # See https://k2-fsa.github.io/sherpa/onnx/pretrained_models/index.html
    # for download links.
    recognizer = sherpa_onnx.OnlineRecognizer.from_transducer(
        tokens=args.tokens,
        encoder=args.encoder,
        decoder=args.decoder,
        joiner=args.joiner,
        num_threads=1,
        sample_rate=16000,
        feature_dim=80,
        decoding_method=args.decoding_method,
        max_active_paths=args.max_active_paths,
        provider=args.provider,
        hotwords_file=args.hotwords_file,
        hotwords_score=args.hotwords_score,
        blank_penalty=args.blank_penalty,
        hr_dict_dir=args.hr_dict_dir,
        hr_rule_fsts=args.hr_rule_fsts,
        hr_lexicon=args.hr_lexicon,
    )
    return recognizer


def main():
    args = get_args()

    devices = sd.query_devices()
    if len(devices) == 0:
        print("No microphone devices found")
        sys.exit(0)

    print(devices)
    default_input_device_idx = sd.default.device[0]
    print(f'Use default device: {devices[default_input_device_idx]["name"]}')

    recognizer = create_recognizer(args)
    print("Started! Please speak")

    # The model is using 16 kHz, we use 48 kHz here to demonstrate that
    # sherpa-onnx will do resampling inside.
    sample_rate = 48000
    samples_per_read = int(0.1 * sample_rate)  # 0.1 second = 100 ms
    last_result = ""
    stream = recognizer.create_stream()
    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)
            stream.accept_waveform(sample_rate, samples)
            while recognizer.is_ready(stream):
                recognizer.decode_stream(stream)
            result = recognizer.get_result(stream)
            if last_result != result:
                last_result = result
                print("\r{}".format(result), end="", flush=True)


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