scripts/tortoise_tts.py

#!/usr/bin/env python3
# AGPL: a notification must be added stating that changes have been made to that file.

import os
import sys
import tempfile
from dataclasses import dataclass
from pathlib import Path
from typing import Literal, Optional

import torch
import torchaudio
from simple_parsing import ArgumentParser, field

from tortoise.api import MODELS_DIR, TextToSpeech
from tortoise.utils.audio import load_audio
from tortoise.utils.diffusion import SAMPLERS
from tortoise.models.vocoder import VocConf


@dataclass
class General:
    """General options"""

    text: str = field(positional=True, nargs="*", metavar="text")
    """Text to speak. If omitted, text is read from stdin."""

    voice: str = field(default="random", alias=["-v"])
    """Selects the voice to use for generation. Use the & character to join two voices together.
    Use a comma to perform inference on multiple voices. Set to "all" to use all available voices.
    Note that multiple voices require the --output-dir option to be set."""

    voices_dir: Optional[str] = field(default=None, alias=["-V"])
    """Path to directory containing extra voices to be loaded. Use a comma to specify multiple directories."""

    preset: Literal["ultra_fast", "fast", "standard", "high_quality"] = field(
        default="fast", alias=["-p"]
    )
    """Which voice quality preset to use."""

    quiet: bool = field(default=False, alias=["-q"])
    """Suppress all output."""

    voicefixer: bool = field(default=True)
    """Enable/Disable voicefixer"""


@dataclass
class Output:
    """Output options"""

    list_voices: bool = field(default=False, alias=["-l"])
    """List available voices and exit."""

    play: bool = field(default=False, alias=["-P"])
    """Play the audio (requires pydub)."""

    output: Optional[Path] = field(default=None, alias=["-o"])
    """Save the audio to a file."""

    output_dir: Path = field(default=Path("results/"), alias=["-O"])
    """Save the audio to a directory as individual segments."""


@dataclass
class MultiOutput:
    """Multi-output options"""

    candidates: int = 1
    """How many output candidates to produce per-voice. Note that only the first candidate is used in the combined output."""

    regenerate: Optional[str] = None
    """Comma-separated list of clip numbers to re-generate."""

    skip_existing: bool = False
    """Set to skip re-generating existing clips."""


@dataclass
class Advanced:
    """Advanced options"""

    produce_debug_state: bool = False
    """Whether or not to produce debug_states in current directory, which can aid in reproducing problems."""

    seed: Optional[int] = None
    """Random seed which can be used to reproduce results."""

    models_dir: str = MODELS_DIR
    """Where to find pretrained model checkpoints. Tortoise automatically downloads these to
    ~/.cache/tortoise/.models, so this should only be specified if you have custom checkpoints."""

    text_split: Optional[str] = None
    """How big chunks to split the text into, in the format <desired_length>,<max_length>."""

    disable_redaction: bool = False
    """Normally text enclosed in brackets are automatically redacted from the spoken output
    (but are still rendered by the model), this can be used for prompt engineering.
    Set this to disable this behavior."""

    device: Optional[str] = None
    """Device to use for inference."""

    batch_size: Optional[int] = None
    """Batch size to use for inference. If omitted, the batch size is set based on available GPU memory."""

    vocoder: Literal["Univnet", "BigVGAN", "BigVGAN_Base"] = "BigVGAN_Base"
    """Pretrained vocoder to be used.
    Univnet - tortoise original
    BigVGAN - 112M model
    BigVGAN_Base - 14M model
    """

    ar_checkpoint: Optional[str] = None
    """Path to a checkpoint to use for the autoregressive model. If omitted, the default checkpoint is used."""

    clvp_checkpoint: Optional[str] = None
    """Path to a checkpoint to use for the CLVP model. If omitted, the default checkpoint is used."""

    diff_checkpoint: Optional[str] = None
    """Path to a checkpoint to use for the diffusion model. If omitted, the default checkpoint is used."""


@dataclass
class Tuning:
    """Tuning options (overrides preset settings)"""

    num_autoregressive_samples: Optional[int] = None
    """Number of samples taken from the autoregressive model, all of which are filtered using CLVP.
    As TorToiSe is a probabilistic model, more samples means a higher probability of creating something "great"."""

    temperature: Optional[float] = None
    """The softmax temperature of the autoregressive model."""

    length_penalty: Optional[float] = None
    """A length penalty applied to the autoregressive decoder. Higher settings causes the model to produce more terse outputs."""

    repetition_penalty: Optional[float] = None
    """A penalty that prevents the autoregressive decoder from repeating itself during decoding.
    Can be used to reduce the incidence of long silences or "uhhhhhhs", etc."""

    top_p: Optional[float] = None
    """P value used in nucleus sampling. 0 to 1. Lower values mean the decoder produces more "likely" (aka boring) outputs."""

    max_mel_tokens: Optional[int] = None
    """Restricts the output length. 1 to 600. Each unit is 1/20 of a second."""

    cvvp_amount: Optional[float] = None
    """How much the CVVP model should influence the output.
    Increasing this can in some cases reduce the likelihood of multiple speakers."""

    diffusion_iterations: Optional[int] = None
    """Number of diffusion steps to perform.  More steps means the network has more chances to iteratively
    refine the output, which should theoretically mean a higher quality output.
    Generally a value above 250 is not noticeably better, however."""

    cond_free: Optional[bool] = None
    """Whether or not to perform conditioning-free diffusion. Conditioning-free diffusion performs two forward passes for
    each diffusion step: one with the outputs of the autoregressive model and one with no conditioning priors. The output
    of the two is blended according to the cond_free_k value below. Conditioning-free diffusion is the real deal, and
    dramatically improves realism."""

    cond_free_k: Optional[float] = None
    """Knob that determines how to balance the conditioning free signal with the conditioning-present signal. [0,inf].
    As cond_free_k increases, the output becomes dominated by the conditioning-free signal.
    Formula is: output=cond_present_output*(cond_free_k+1)-cond_absenct_output*cond_free_k"""

    diffusion_temperature: Optional[float] = None
    """Controls the variance of the noise fed into the diffusion model. [0,1]. Values at 0
    are the "mean" prediction of the diffusion network and will sound bland and smeared."""


@dataclass
class Speed:
    """New/speed options"""

    low_vram: bool = False
    """re-enable default offloading behaviour of tortoise"""

    half: bool = False
    """enable autocast to half precision for autoregressive model"""

    no_cache: bool = False
    """disable kv_cache usage. This should really only be used if you are very low on vram."""

    sampler: Optional[str] = field(default=None, choices=SAMPLERS)
    """override the sampler used for diffusion (default depends on --preset)"""

    original_tortoise: bool = False
    """ensure results are identical to original tortoise-tts repo"""


if __name__ == "__main__":
    parser = ArgumentParser(
        description="TorToiSe is a text-to-speech program that is capable of synthesizing speech "
        "in multiple voices with realistic prosody and intonation."
    )
    # bugs out for some reason
    # parser.add_argument(
    #     "--web",
    #     action="store_true",
    #     help="launch the webui (doesn't pass it the other arguments)",
    # )
    parser.add_arguments(General, "general")
    parser.add_arguments(Output, "output")
    parser.add_arguments(MultiOutput, "multi_output")
    parser.add_arguments(Advanced, "advanced")
    parser.add_arguments(Tuning, "tuning")
    parser.add_arguments(Speed, "speed")

    usage_examples = f"""
    Examples:

    Read text using random voice and place it in a file:

        {parser.prog} -o hello.wav "Hello, how are you?"

    Read text from stdin and play it using the tom voice:

        echo "Say it like you mean it!" | {parser.prog} -P -v tom

    Read a text file using multiple voices and save the audio clips to a directory:

        {parser.prog} -O /tmp/tts-results -v tom,emma <textfile.txt
    """

    # show usage even when Ctrl+C is pressed early
    try:
        args = parser.parse_args()
    except SystemExit as e:
        if e.code == 0:
            print(usage_examples)
        sys.exit(e.code)
    # bugs out for some reason
    # if args.web:
    #     from importlib import import_module
    #     app = import_module("app")
    #     sys.exit(app.main())

    from tortoise.inference import (
        check_pydub,
        get_all_voices,
        get_seed,
        parse_multiarg_text,
        parse_voice_str,
        split_text,
        validate_output_dir,
        voice_loader,
        save_gen_with_voicefix
    )

    # get voices
    all_voices, extra_voice_dirs = get_all_voices(args.general.voices_dir)
    if args.output.list_voices:
        for v in all_voices:
            print(v)
        sys.exit(0)
    selected_voices = parse_voice_str(args.general.voice, all_voices)
    voice_generator = voice_loader(selected_voices, extra_voice_dirs)

    # parse text
    if not args.general.text:
        print("reading text from stdin!")
    text = parse_multiarg_text(args.general.text)
    texts = split_text(text, args.advanced.text_split)

    output_dir = validate_output_dir(
        args.output.output_dir, selected_voices, args.multi_output.candidates
    )

    # error out early if pydub isn't installed
    pydub = check_pydub(args.output.play)

    seed = get_seed(args.advanced.seed)
    verbose = not args.general.quiet

    vocoder = getattr(VocConf, args.advanced.vocoder)
    if verbose:
        print("Loading tts...")
    tts = TextToSpeech(
        models_dir=args.advanced.models_dir,
        enable_redaction=not args.advanced.disable_redaction,
        device=args.advanced.device,
        autoregressive_batch_size=args.advanced.batch_size,
        high_vram=not args.speed.low_vram,
        kv_cache=not args.speed.no_cache,
        ar_checkpoint=args.advanced.ar_checkpoint,
        clvp_checkpoint=args.advanced.clvp_checkpoint,
        diff_checkpoint=args.advanced.diff_checkpoint,
        vocoder=vocoder,
    )

    gen_settings = {
        "use_deterministic_seed": seed,
        "verbose": verbose,
        "k": args.multi_output.candidates,
        "preset": args.general.preset,
    }
    tuning_options = [
        "num_autoregressive_samples",
        "temperature",
        "length_penalty",
        "repetition_penalty",
        "top_p",
        "max_mel_tokens",
        "cvvp_amount",
        "diffusion_iterations",
        "cond_free",
        "cond_free_k",
        "diffusion_temperature",
    ]
    for option in tuning_options:
        if getattr(args.tuning, option) is not None:
            gen_settings[option] = getattr(args.tuning, option)

    speed_options = [
        "sampler",
        "original_tortoise",
        "half",
    ]
    for option in speed_options:
        if getattr(args.speed, option) is not None:
            gen_settings[option] = getattr(args.speed, option)

    total_clips = len(texts) * len(selected_voices)
    regenerate_clips = (
        [int(x) for x in args.multi_output.regenerate.split(",")]
        if args.multi_output.regenerate
        else None
    )
    for voice_idx, (voice, voice_samples, conditioning_latents) in enumerate(
        voice_generator
    ):
        audio_parts = []
        for text_idx, text in enumerate(texts):
            clip_name = f'{"-".join(voice)}_{text_idx:02d}'
            if args.output.output_dir:
                first_clip = os.path.join(args.output.output_dir, f"{clip_name}_00.wav")
                if (
                    args.multi_output.skip_existing
                    or (regenerate_clips and text_idx not in regenerate_clips)
                ) and os.path.exists(first_clip):
                    audio_parts.append(load_audio(first_clip, 24000))
                    if verbose:
                        print(f"Skipping {clip_name}")
                    continue
            if verbose:
                print(
                    f"Rendering {clip_name} ({(voice_idx * len(texts) + text_idx + 1)} of {total_clips})..."
                )
                print("  " + text)
            gen = tts.tts_with_preset(
                text,
                voice_samples=voice_samples,
                conditioning_latents=conditioning_latents,
                **gen_settings,
            )
            gen = gen if args.multi_output.candidates > 1 else [gen]
            for candidate_idx, audio in enumerate(gen):
                audio = audio.squeeze(0).cpu()
                if candidate_idx == 0:
                    audio_parts.append(audio)
                if args.output.output_dir:
                    filename = f"{clip_name}_{candidate_idx:02d}.wav"
                    save_gen_with_voicefix(audio, os.path.join(args.output.output_dir, filename), squeeze=False, voicefixer=args.general.voicefixer)

        audio = torch.cat(audio_parts, dim=-1)
        if args.output.output_dir:
            filename = f'{"-".join(voice)}_combined.wav'
            save_gen_with_voicefix(
                audio,
                os.path.join(args.output.output_dir, filename),
                squeeze=False,
                voicefixer=args.general.voicefixer,
            )
        elif args.output.output:
            filename = args.output.output or os.tmp
            save_gen_with_voicefix(audio, filename, squeeze=False, voicefixer=args.general.voicefixer)
        elif args.output.play:
            print("WARNING: cannot use voicefixer with --play")
            f = tempfile.NamedTemporaryFile(suffix=".wav", delete=True)
            torchaudio.save(f.name, audio, 24000)
            pydub.playback.play(pydub.AudioSegment.from_wav(f.name))

        if args.advanced.produce_debug_state:
            os.makedirs("debug_states", exist_ok=True)
            dbg_state = (seed, texts, voice_samples, conditioning_latents, args)
            torch.save(
                dbg_state, os.path.join("debug_states", f'debug_{"-".join(voice)}.pth')
            )