enginex_bi_series-sherpa-onnx/python-api-examples/offline-tts.py

#!/usr/bin/env python3
#
# Copyright (c)  2023  Xiaomi Corporation

"""
This file demonstrates how to use sherpa-onnx Python API to generate audio
from text, i.e., text-to-speech.


Different from ./offline-tts-play.py, this file does not play back the
generated audio.

Usage:

Example (1/2)

wget https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-en_US-amy-low.tar.bz2
tar xf vits-piper-en_US-amy-low.tar.bz2

python3 ./python-api-examples/offline-tts.py \
 --vits-model=./vits-piper-en_US-amy-low/en_US-amy-low.onnx \
 --vits-tokens=./vits-piper-en_US-amy-low/tokens.txt \
 --vits-data-dir=./vits-piper-en_US-amy-low/espeak-ng-data \
 --output-filename=./generated.wav \
 "Today as always, men fall into two groups: slaves and free men. Whoever does not have two-thirds of his day for himself, is a slave, whatever he may be: a statesman, a businessman, an official, or a scholar."

Example (2/2)

wget https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-zh-aishell3.tar.bz2
tar xvf vits-zh-aishell3.tar.bz2

python3 ./python-api-examples/offline-tts.py \
 --vits-model=./vits-aishell3.onnx \
 --vits-lexicon=./lexicon.txt \
 --vits-tokens=./tokens.txt \
 --tts-rule-fsts=./rule.fst \
 --sid=21 \
 --output-filename=./liubei-21.wav \
 "勿以恶小而为之，勿以善小而不为。惟贤惟德，能服于人。122334"

You can find more models at
https://github.com/k2-fsa/sherpa-onnx/releases/tag/tts-models

Please see
https://k2-fsa.github.io/sherpa/onnx/tts/index.html
for details.

"""

import argparse
import time

import sherpa_onnx
import soundfile as sf


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

    parser.add_argument(
        "--vits-model",
        type=str,
        help="Path to vits model.onnx",
    )

    parser.add_argument(
        "--vits-lexicon",
        type=str,
        default="",
        help="Path to lexicon.txt",
    )

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

    parser.add_argument(
        "--vits-data-dir",
        type=str,
        default="",
        help="""Path to the dict director of espeak-ng. If it is specified,
        --vits-lexicon and --vits-tokens are ignored""",
    )

    parser.add_argument(
        "--tts-rule-fsts",
        type=str,
        default="",
        help="Path to rule.fst",
    )

    parser.add_argument(
        "--max-num-sentences",
        type=int,
        default=2,
        help="""Max number of sentences in a batch to avoid OOM if the input
        text is very long. Set it to -1 to process all the sentences in a
        single batch. A smaller value does not mean it is slower compared
        to a larger one on CPU.
        """,
    )

    parser.add_argument(
        "--output-filename",
        type=str,
        default="./generated.wav",
        help="Path to save generated wave",
    )

    parser.add_argument(
        "--sid",
        type=int,
        default=0,
        help="""Speaker ID. Used only for multi-speaker models, e.g.
        models trained using the VCTK dataset. Not used for single-speaker
        models, e.g., models trained using the LJ speech dataset.
        """,
    )

    parser.add_argument(
        "--debug",
        type=bool,
        default=False,
        help="True to show debug messages",
    )

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

    parser.add_argument(
        "--num-threads",
        type=int,
        default=1,
        help="Number of threads for neural network computation",
    )

    parser.add_argument(
        "--speed",
        type=float,
        default=1.0,
        help="Speech speed. Larger->faster; smaller->slower",
    )

    parser.add_argument(
        "text",
        type=str,
        help="The input text to generate audio for",
    )

    return parser.parse_args()


def main():
    args = get_args()
    print(args)

    tts_config = sherpa_onnx.OfflineTtsConfig(
        model=sherpa_onnx.OfflineTtsModelConfig(
            vits=sherpa_onnx.OfflineTtsVitsModelConfig(
                model=args.vits_model,
                lexicon=args.vits_lexicon,
                data_dir=args.vits_data_dir,
                tokens=args.vits_tokens,
            ),
            provider=args.provider,
            debug=args.debug,
            num_threads=args.num_threads,
        ),
        rule_fsts=args.tts_rule_fsts,
        max_num_sentences=args.max_num_sentences,
    )
    if not tts_config.validate():
        raise ValueError("Please check your config")

    tts = sherpa_onnx.OfflineTts(tts_config)

    start = time.time()
    audio = tts.generate(args.text, sid=args.sid, speed=args.speed)
    end = time.time()

    if len(audio.samples) == 0:
        print("Error in generating audios. Please read previous error messages.")
        return

    elapsed_seconds = end - start
    audio_duration = len(audio.samples) / audio.sample_rate
    real_time_factor = elapsed_seconds / audio_duration

    sf.write(
        args.output_filename,
        audio.samples,
        samplerate=audio.sample_rate,
        subtype="PCM_16",
    )
    print(f"Saved to {args.output_filename}")
    print(f"The text is '{args.text}'")
    print(f"Elapsed seconds: {elapsed_seconds:.3f}")
    print(f"Audio duration in seconds: {audio_duration:.3f}")
    print(f"RTF: {elapsed_seconds:.3f}/{audio_duration:.3f} = {real_time_factor:.3f}")


if __name__ == "__main__":
    main()