Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

2.8 KB
Raw Blame
# SPDX-License-Identifier: AGPL-3.0-only
# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0

"""Which Audio page workflows (Speak, Clone, Edit, Convert, Music, Separate, Transcribe) a model or
clip belongs to; mirrors studio/frontend/src/features/audio/workflows.ts. Reads audio.cpp
constants; audio_cpp_models never imports it."""

from __future__ import annotations

from typing import Optional

from core.inference.audio_cpp_models import (
    AUDIO_CPP_MUSIC_AUDIO_TYPE,
    AUDIO_CPP_SEP_AUDIO_TYPE,
    HUB_TASKS,
)

AUDIO_WORKFLOW_IDS = ("speak", "clone", "edit", "convert", "music", "separate", "transcribe")

# The page names people see in the sidebar.
AUDIO_WORKFLOW_LABELS = {
    "speak": "Text to Speech",
    "clone": "Clone",
    "edit": "Edit",
    "convert": "Convert",
    "music": "Music",
    "separate": "Separate",
    "transcribe": "Transcribe",
}

# Generation audio types that make music; every other generation audio type speaks.
MUSIC_AUDIO_TYPES = frozenset(("minimax_music3", AUDIO_CPP_MUSIC_AUDIO_TYPE))

# Audio types of a loaded model that transcribes rather than generates.
_TRANSCRIBE_AUDIO_TYPES = frozenset(("whisper",))


def workflow_for_audio_type(audio_type: Optional[str]) -> str:
    """The generation workflow (``speak`` or ``music``) that a model or clip of this type serves."""
    return "music" if audio_type in MUSIC_AUDIO_TYPES else "speak"


def status_audio_workflows(is_audio: bool, audio_type: Optional[str]) -> list[str]:
    """Workflows the loaded model serves; empty for a model that is not an audio model."""
    if not is_audio:
        return []
    if audio_type in _TRANSCRIBE_AUDIO_TYPES:
        return ["transcribe"]
    if audio_type == AUDIO_CPP_SEP_AUDIO_TYPE:
        return ["separate"]
    return [workflow_for_audio_type(audio_type)]


def inventory_audio_workflows(
    task: Optional[str], audio_type: Optional[str]
) -> Optional[list[str]]:
    """The pipeline task decides first, as music rows lack audio_type; separation is the exception."""
    if audio_type == AUDIO_CPP_SEP_AUDIO_TYPE:
        return ["separate"]
    if task == HUB_TASKS["music"]:
        return ["music"]
    if task == HUB_TASKS["asr"]:
        return ["transcribe"]
    if task == HUB_TASKS["tts"]:
        return [workflow_for_audio_type(audio_type)]
    return None


def pages_sentence(workflows) -> str:
    """``It runs on Text to Speech (speak) and Clone (clone).``, for an error naming a model's pages."""
    runs = set(workflows or ())
    names = [f"{AUDIO_WORKFLOW_LABELS[wf]} ({wf})" for wf in AUDIO_WORKFLOW_IDS if wf in runs]
    if not names:
        return "Unsloth Studio has no Audio page for it yet."
    pages = names[0] if len(names) == 1 else ", ".join(names[:-1]) + " and " + names[-1]
    return f"It runs on {pages}."