model.py

# Copyright      2022-2023  Xiaomi Corp.        (authors: Fangjun Kuang)
#
# See LICENSE for clarification regarding multiple authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from functools import lru_cache

import sherpa_onnx
from huggingface_hub import hf_hub_download


def get_file(
    repo_id: str,
    filename: str,
    subfolder: str = ".",
) -> str:
    model_filename = hf_hub_download(
        repo_id=repo_id,
        filename=filename,
        subfolder=subfolder,
    )
    return model_filename


@lru_cache(maxsize=10)
def _get_vits_vctk(repo_id: str, speed: float) -> sherpa_onnx.OfflineTts:
    assert repo_id == "csukuangfj/vits-vctk"

    model = get_file(
        repo_id=repo_id,
        filename="vits-vctk.onnx",
        subfolder=".",
    )

    lexicon = get_file(
        repo_id=repo_id,
        filename="lexicon.txt",
        subfolder=".",
    )

    tokens = get_file(
        repo_id=repo_id,
        filename="tokens.txt",
        subfolder=".",
    )

    tts_config = sherpa_onnx.OfflineTtsConfig(
        model=sherpa_onnx.OfflineTtsModelConfig(
            vits=sherpa_onnx.OfflineTtsVitsModelConfig(
                model=model,
                lexicon=lexicon,
                tokens=tokens,
                length_scale=1.0 / speed,
            ),
            provider="cpu",
            debug=True,
            num_threads=2,
        )
    )
    tts = sherpa_onnx.OfflineTts(tts_config)

    return tts


@lru_cache(maxsize=10)
def _get_vits_ljs(repo_id: str, speed: float) -> sherpa_onnx.OfflineTts:
    assert repo_id == "csukuangfj/vits-ljs"

    model = get_file(
        repo_id=repo_id,
        filename="vits-ljs.onnx",
        subfolder=".",
    )

    lexicon = get_file(
        repo_id=repo_id,
        filename="lexicon.txt",
        subfolder=".",
    )

    tokens = get_file(
        repo_id=repo_id,
        filename="tokens.txt",
        subfolder=".",
    )

    tts_config = sherpa_onnx.OfflineTtsConfig(
        model=sherpa_onnx.OfflineTtsModelConfig(
            vits=sherpa_onnx.OfflineTtsVitsModelConfig(
                model=model,
                lexicon=lexicon,
                tokens=tokens,
                length_scale=1.0 / speed,
            ),
            provider="cpu",
            debug=True,
            num_threads=2,
        )
    )
    tts = sherpa_onnx.OfflineTts(tts_config)

    return tts


@lru_cache(maxsize=10)
def _get_vits_piper(repo_id: str, speed: float) -> sherpa_onnx.OfflineTts:
    data_dir = "/tmp/espeak-ng-data"
    if "coqui" in repo_id or "vits-mms" in repo_id:
        name = "model"
    elif "piper" in repo_id:
        n = len("vits-piper-")
        name = repo_id.split("/")[1][n:]
    elif "mimic3" in repo_id:
        n = len("vits-mimic3-")
        name = repo_id.split("/")[1][n:]
    else:
        raise ValueError(f"Unsupported {repo_id}")

    if "vits-coqui-uk-mai" in repo_id or "vits-mms" in repo_id:
        data_dir = ""

    model = get_file(
        repo_id=repo_id,
        filename=f"{name}.onnx",
        subfolder=".",
    )

    tokens = get_file(
        repo_id=repo_id,
        filename="tokens.txt",
        subfolder=".",
    )

    tts_config = sherpa_onnx.OfflineTtsConfig(
        model=sherpa_onnx.OfflineTtsModelConfig(
            vits=sherpa_onnx.OfflineTtsVitsModelConfig(
                model=model,
                lexicon="",
                data_dir=data_dir,
                tokens=tokens,
                length_scale=1.0 / speed,
            ),
            provider="cpu",
            debug=True,
            num_threads=2,
        )
    )
    tts = sherpa_onnx.OfflineTts(tts_config)

    return tts


@lru_cache(maxsize=10)
def _get_vits_mms(repo_id: str, speed: float) -> sherpa_onnx.OfflineTts:
    return _get_vits_piper(repo_id, speed)


@lru_cache(maxsize=10)
def _get_vits_zh_aishell3(repo_id: str, speed: float) -> sherpa_onnx.OfflineTts:
    assert repo_id == "csukuangfj/vits-zh-aishell3"

    model = get_file(
        repo_id=repo_id,
        filename="vits-aishell3.onnx",
        subfolder=".",
    )

    lexicon = get_file(
        repo_id=repo_id,
        filename="lexicon.txt",
        subfolder=".",
    )

    tokens = get_file(
        repo_id=repo_id,
        filename="tokens.txt",
        subfolder=".",
    )

    rule_fst = get_file(
        repo_id=repo_id,
        filename="rule.fst",
        subfolder=".",
    )

    tts_config = sherpa_onnx.OfflineTtsConfig(
        model=sherpa_onnx.OfflineTtsModelConfig(
            vits=sherpa_onnx.OfflineTtsVitsModelConfig(
                model=model,
                lexicon=lexicon,
                tokens=tokens,
                length_scale=1.0 / speed,
            ),
            provider="cpu",
            debug=True,
            num_threads=2,
        ),
        rule_fsts=rule_fst,
    )
    tts = sherpa_onnx.OfflineTts(tts_config)

    return tts


@lru_cache(maxsize=10)
def _get_vits_hf(repo_id: str, speed: float) -> sherpa_onnx.OfflineTts:
    if "fanchen" in repo_id or "vits-cantonese-hf-xiaomaiiwn" in repo_id:
        model = repo_id.split("/")[-1]
    else:
        model = repo_id.split("-")[-1]

    model = get_file(
        repo_id=repo_id,
        filename=f"{model}.onnx",
        subfolder=".",
    )

    lexicon = get_file(
        repo_id=repo_id,
        filename="lexicon.txt",
        subfolder=".",
    )

    tokens = get_file(
        repo_id=repo_id,
        filename="tokens.txt",
        subfolder=".",
    )

    rule_fst = get_file(
        repo_id=repo_id,
        filename="rule.fst",
        subfolder=".",
    )

    tts_config = sherpa_onnx.OfflineTtsConfig(
        model=sherpa_onnx.OfflineTtsModelConfig(
            vits=sherpa_onnx.OfflineTtsVitsModelConfig(
                model=model,
                lexicon=lexicon,
                tokens=tokens,
                length_scale=1.0 / speed,
            ),
            provider="cpu",
            debug=True,
            num_threads=2,
        ),
        rule_fsts=rule_fst,
    )
    tts = sherpa_onnx.OfflineTts(tts_config)

    return tts


@lru_cache(maxsize=10)
def get_pretrained_model(repo_id: str, speed: float) -> sherpa_onnx.OfflineTts:
    if repo_id in spanish_models:
        return spanish_models[repo_id](repo_id, speed)
    else:
        raise ValueError(f"Unsupported repo_id: {repo_id}")

spanish_models = {
    "csukuangfj/vits-piper-es_ES-sharvard-medium": _get_vits_piper,  # 2 speakers
}

language_to_models = {
    "Spanish": list(spanish_models.keys()),
}