diff --git a/ARCHITECTURE.md b/ARCHITECTURE.md index 4e085a9..234febd 100644 --- a/ARCHITECTURE.md +++ b/ARCHITECTURE.md @@ -237,6 +237,10 @@ covered by the `styles.css` manifest. character's language, speed and folded sampling parameters to its `defaults` (only values that differ from the node's own defaults, which it reads from `/object_info`; needs Anomalous_TTS interface 11). + `ui_tts_pronunciation.js` edits a GPT-SoVITS character's pronunciation table + (the node's `replace` setting) from its card; each row can be heard as it reads + now and as replaced, through a Preview Audio run, and saving keeps every other + settings field. `audio_tts_run.js` queues one API prompt with this page's client id and follows it (websocket status, `/history` result, cancel = queue delete or targeted `/interrupt`); it never touches the canvas. diff --git a/CHANGELOG.md b/CHANGELOG.md index ef1f44d..3f6f75b 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -5,6 +5,7 @@ ### 🎙️ Audio & Voice Studio (音频与声音工作台) - **Voice library for ComfyUI-F5-TTS (F5-TTS 声音库)**: follows F5-TTS's own layout — `Character.wav` is the main voice and `Character..wav` are its emotion variants (used as `{emotion}` tags). Voices can be previewed, uploaded with their reference text (with one-click romanization for Japanese/Korean text), and dragged onto a TTS node; the node's own path spelling is used on every OS. - **Generate right in the Script Director (剧本台直接生成)**: for GPT-SoVITS characters, **Generate** runs the script in ComfyUI without touching the canvas (it waits in the queue while ComfyUI is busy), plays the result and saves it to `output/audio//`. **↻** on a line card gives just that line another take while the others come from cache; **New take of all** starts over with a new seed. Language, speed and the sampling parameters (folded under **Advanced**, with a hint on each) start from the character's saved defaults; when a take sounds right, **Save these settings as this character's default** keeps them for next time. The script drawer is wider and its footer scrolls, so the line cards keep room. Needs Anomalous_TTS interface 11. +- **Pronunciation table (读音表)**: **Pronunciation** on a GPT-SoVITS character card lists what the script says and what the character should read instead (`C站 → 西站`); each line can be heard as the character reads it now and as replaced, without filling the audio gallery. - **Audio page tabs (音频页顶部标签)**: the "Audio Workflows" tab is gone from the audio page; generating now happens in the Script Director, and saved workflows stay under **Workflows** on the image side. - **Script Director (剧本台)**: paste a script for one character; each sentence becomes a card with emotion chips, inline editing, split at cursor, merge and delete. The finished script is pushed to the selected F5-TTS node or dragged onto one, with the tags written for you. - **Audio gallery (音频库)**: lists generated audio with its speech text, voice and seed read from the file itself, supports search, and can save unsaved previews into `output/audio/`. diff --git a/docs/guides/audio-studio.md b/docs/guides/audio-studio.md index a69f4d2..5a34c2c 100644 --- a/docs/guides/audio-studio.md +++ b/docs/guides/audio-studio.md @@ -124,6 +124,14 @@ GPT-SoVITS 角色在剧本台里多一块 **生成** 区： 需要 Anomalous_TTS 接口版本 11 以上；旧版本时这里会提示更新，仍然可以推送到节点、在画布上运行。 +### GPT-SoVITS：某个词总是读错 + +角色卡片上的 **读音** 打开这个角色的读音表：左边写剧本里的字，右边写让它实际读的字，例如 `C站 → 西站`、`粘进 → 沾进`。剧本和字幕照常写，只影响念出来的声音。 + +- 每一行的 **现在** 听这个角色现在怎么读左边的字，**改后** 听右边的字，对比着改。试听不会存进音频库。 +- 日语模型说中文时特别有用：它发不好的音（选、全这类 ü，模型、行这类后鼻音）换成好读的同音字，或在前面加个字；英文字母写成读音相近的汉字。 +- 长的先匹配，只替换一遍。只对这个角色生效。 + ## 4. 生成和找回结果 - F5-TTS：把剧本推送到画布上的节点后运行。GPT-SoVITS 也可以在剧本台里直接生成（见上）。 diff --git a/web/modules/audio_script.js b/web/modules/audio_script.js index 8222c92..bef504d 100644 --- a/web/modules/audio_script.js +++ b/web/modules/audio_script.js @@ -143,19 +143,19 @@ export function ttsOutputPrefix(character) { export const TTS_NODE_CLASS = 'AnomalousTTS_CharacterSpeech'; /** - * ComfyUI API prompt for one GPT-SoVITS script: the character node, then Save Audio. + * ComfyUI API prompt for one GPT-SoVITS script: the character node, then Save Audio + * (`preview: true`: Preview Audio, a temp file that stays out of the audio gallery). * `inputs` are more node inputs by name (speed, top_k …); left out, the node's defaults apply. */ -export function buildTtsPrompt({ character, speech, seed, language = 'auto', ...inputs }) { +export function buildTtsPrompt({ character, speech, seed, language = 'auto', preview = false, ...inputs }) { return { 1: { class_type: TTS_NODE_CLASS, inputs: { character, text: speech, seed, language, ...inputs }, }, - 2: { - class_type: 'SaveAudio', - inputs: { audio: ['1', 0], filename_prefix: ttsOutputPrefix(character) }, - }, + 2: preview + ? { class_type: 'PreviewAudio', inputs: { audio: ['1', 0] } } + : { class_type: 'SaveAudio', inputs: { audio: ['1', 0], filename_prefix: ttsOutputPrefix(character) } }, }; } diff --git a/web/modules/locales.js b/web/modules/locales.js index 60b5f9a..ee961a1 100644 --- a/web/modules/locales.js +++ b/web/modules/locales.js @@ -1333,6 +1333,20 @@ export const i18n = { audioDragCardHintGptSovits: "拖标题栏 → GPT-SoVITS 节点 · 点 {标签} 复制", audioDragCharacterTitleGptSovits: "拖到 GPT-SoVITS 节点：角色设为 {character}。情绪在台词里用 {标签} 切换。", audioDropReject_otherEngine: "这是 {voiceEngine} 的声音，{node} 节点用不了，请拖到 {voiceEngine} 的节点上", + ttsPronounceOpen: '读音', + ttsPronounceTitle: '读音：{name}', + ttsPronounceSubtitle: '剧本里写左边的字，这个角色就读右边的字。剧本和字幕照常写，只影响念出来的声音', + ttsPronounceTips: '长的先匹配，只替换一遍。日语模型说中文时：发不好的音（选、全这类 ü，模型、行这类后鼻音）换成好读的同音字，或在前面加个字；英文字母写成读音相近的汉字（C站 → 西站）；多音字读错就换成同音字（粘进 → 沾进）。右边留空表示不读。', + ttsPronounceLanguage: '试听语言', + ttsPronounceFromPlaceholder: '剧本里写的', + ttsPronounceToPlaceholder: '让角色读成', + ttsPronounceNow: '现在', + ttsPronounceNowHint: '听这个角色现在怎么读左边的字（已保存的读音表会生效）', + ttsPronounceAfter: '改后', + ttsPronounceAfterHint: '听右边的字读出来是什么样', + ttsPronounceAdd: '+ 添加一条', + ttsPronounceFromMissing: '有一条只写了读法（{to}），左边要写剧本里的字', + ttsPronounceDuplicate: '“{from}”写了两遍，请合并成一条', ttsEditorOpen: "编辑情绪", ttsEditorTitle: "编辑情绪 · {name}", ttsEditorSubtitle: "为主声音和每种情绪各挑一条参考音频（3~10 秒、有台词最好）。保存后写进角色文件夹里的 anomalous_tts.json。", @@ -3108,6 +3122,20 @@ export const i18n = { audioDragCardHintGptSovits: "Drag the header → GPT-SoVITS node · click a {tag} to copy", audioDragCharacterTitleGptSovits: "Drop onto a GPT-SoVITS node to set its character to {character}. Switch emotions with {tags} in the text.", audioDropReject_otherEngine: "This is a {voiceEngine} voice; the {node} node cannot use it. Drop it on the {voiceEngine} node", + ttsPronounceOpen: 'Pronunciation', + ttsPronounceTitle: 'Pronunciation: {name}', + ttsPronounceSubtitle: 'When the script says the left text, this character reads the right text. Scripts and subtitles stay as written; only the voice changes', + ttsPronounceTips: 'Longest match first, replaced once. A Japanese model speaking Chinese: swap sounds it cannot say (ü as in 选, -ng as in 模型) for easier same-sounding characters, or put a character before the word; write Latin letters as Chinese characters that sound alike (C站 → 西站); fix a misread polyphone with a same-sounding character (粘进 → 沾进). Leave the right side empty to skip the text.', + ttsPronounceLanguage: 'Preview language', + ttsPronounceFromPlaceholder: 'Text in the script', + ttsPronounceToPlaceholder: 'Read it as', + ttsPronounceNow: 'Now', + ttsPronounceNowHint: 'Hear how this character reads the left text now (the saved table applies)', + ttsPronounceAfter: 'After', + ttsPronounceAfterHint: 'Hear the right text read aloud', + ttsPronounceAdd: '+ Add a line', + ttsPronounceFromMissing: 'A line has only a reading ({to}); write the script text on the left', + ttsPronounceDuplicate: '"{from}" is listed twice; keep one line', ttsEditorOpen: "Edit emotions", ttsEditorTitle: "Edit emotions · {name}", ttsEditorSubtitle: "Pick one reference clip (3–10 s, with its line) for the main voice and each emotion. Saved to anomalous_tts.json in the character folder.", diff --git a/web/modules/ui_audio_studio.js b/web/modules/ui_audio_studio.js index 98f38b8..46d8c54 100644 --- a/web/modules/ui_audio_studio.js +++ b/web/modules/ui_audio_studio.js @@ -6,6 +6,7 @@ import { getActiveAudioFilter, renderAudioSidebar, setActiveAudioFilter } from ' import { bindVoiceDrag } from './audio_voice_drag.js'; import { AUDIO_ENGINES, detectEngines, engineById, engineTargetLabels, getStoredEngine, invalidateEngineCache, loadEngine, loadGptSovitsStatus, pickEngine, setStoredEngine } from './audio_engines.js'; import { openGptSovitsEditor } from './ui_audio_tts_editor.js'; +import { openPronunciationEditor } from './ui_tts_pronunciation.js'; import { bindTtsFileDrop } from './ui_tts_file_drop.js'; import { openTtsImport } from './ui_tts_import.js'; import { @@ -39,6 +40,7 @@ const SVG = { SEARCH: ``, SCRIPT: ``, PLUS: ``, + PRONOUNCE: ``, EDIT: ``, REFRESH: ``, }; @@ -251,7 +253,9 @@ function renderCharacterCard(group, owner, { onChanged, canImport = false } = {} if (isTts && group.raw && !group.raw.error) { const edit = createToolButton(SVG.EDIT, t('ttsEditorOpen'), 'is-compact'); edit.onclick = () => openGptSovitsEditor(group, { onSaved: () => onChanged?.() }); - headerRight.append(edit); + const pronounce = createToolButton(SVG.PRONOUNCE, t('ttsPronounceOpen'), 'is-compact'); + pronounce.onclick = () => openPronunciationEditor(group, { onSaved: () => onChanged?.() }); + headerRight.append(edit, pronounce); if (canImport) { const add = createToolButton(SVG.PLUS, t('ttsImportAddOpen'), 'is-compact'); add.onclick = () => openTtsImport({ target: group.character, onDone: () => onChanged?.() }); diff --git a/web/modules/ui_tts_pronunciation.js b/web/modules/ui_tts_pronunciation.js new file mode 100644 index 0000000..858510e --- /dev/null +++ b/web/modules/ui_tts_pronunciation.js @@ -0,0 +1,201 @@ +import { t } from './interface_settings.js'; +import { createViewScope } from './ui_lifecycle.js'; +import { anomalousAlert } from './ui_dialog.js'; +import { buildTtsPrompt } from './audio_script.js'; +import { startPromptJob } from './audio_tts_run.js'; +import { fetchGptSovitsCharacter, saveGptSovitsSettings } from './audio_engines.js'; + +/** + * GPT-SoVITS pronunciation table: "what the script says -> what this character + * reads" (the node's `replace` setting, applied before synthesis). Each row can + * be heard both ways through a short Preview Audio run (audio_tts_run.js): + * "now" sends the original through the node, so the saved table applies; "after" + * sends the replacement. Saving keeps every other settings field. + */ + +let activeScope = null; +const LANGUAGES = ['auto', 'zh', 'ja', 'en']; + +function el(tag, className, text) { + const node = document.createElement(tag); + if (className) node.className = className; + if (text !== undefined) node.textContent = text; + return node; +} + +function button(className, label, onClick, title) { + const btn = el('button', className, label); + btn.type = 'button'; + if (title) { btn.title = title; btn.setAttribute('aria-label', title); } + btn.onclick = onClick; + return btn; +} + +function textInput(className, value, placeholder) { + const input = el('input', className); + input.type = 'text'; + input.placeholder = placeholder; + input.value = value || ''; + return input; +} + +/** First problem as a message, or '' when the rows can be saved. */ +export function validatePronunciationRows(rows) { + const seen = new Set(); + for (const row of rows) { + if (!row.from.trim()) return t('ttsPronounceFromMissing', { to: row.to }); + if (seen.has(row.from)) return t('ttsPronounceDuplicate', { from: row.from }); + seen.add(row.from); + } + return ''; +} + +/** + * Open the table for one GPT-SoVITS voice group (audio_engines.gptSovitsGroups shape). + * `onSaved(group)` receives the refreshed group after the node accepted the settings. + */ +export function openPronunciationEditor(group, { onSaved } = {}) { + activeScope?.dispose(); + const scope = createViewScope(); + activeScope = scope; + const raw = group.raw || {}; + const name = raw.name || group.character; + let settings = raw.settings && typeof raw.settings === 'object' ? raw.settings : {}; + let loaded = false; + let job = null; + const player = new Audio(); + const seed = Math.floor(Math.random() * 2 ** 31); + + const overlay = el('div', 'anomalous-voice-modal-overlay'); + const modal = el('div', 'anomalous-voice-modal anomalous-tts-editor anomalous-tts-pronounce'); + modal.setAttribute('role', 'dialog'); + modal.setAttribute('aria-modal', 'true'); + const close = () => scope.dispose(); + scope.onDispose(() => { + job?.cancel(); + player.pause(); + overlay.remove(); + if (activeScope === scope) activeScope = null; + }); + overlay.onclick = (e) => { if (e.target === overlay) close(); }; + scope.listen(window, 'keydown', (e) => { + if (e.key !== 'Escape' || document.querySelector('.anomalous-dialog-overlay')) return; + e.stopPropagation(); + close(); + }, true); + + const header = el('div', 'anomalous-voice-modal-header'); + const heading = el('div', 'anomalous-voice-modal-heading'); + heading.append( + el('div', 'anomalous-voice-modal-title', t('ttsPronounceTitle', { name })), + el('div', 'anomalous-voice-modal-subtitle', t('ttsPronounceSubtitle')), + ); + header.append(heading, button('anomalous-voice-modal-close', '×', close, t('close'))); + + const languageBar = el('label', 'anomalous-tts-pronounce-language'); + const language = el('select', 'anomalous-sd-run-select'); + for (const code of LANGUAGES) { + const option = el('option', '', t(`scriptRunLang_${code}`)); + option.value = code; + language.appendChild(option); + } + language.value = LANGUAGES.includes(settings.defaults?.language) ? settings.defaults.language : 'auto'; + languageBar.append(el('span', '', t('ttsPronounceLanguage')), language); + + /** Play `text` through the node (Preview Audio); one preview at a time. */ + const preview = async (text, btn) => { + if (!text.trim() || job) return; + player.pause(); + const label = btn.textContent; + btn.textContent = '…'; + btn.classList.add('is-busy'); + job = startPromptJob(buildTtsPrompt({ character: name, speech: text.trim(), seed, language: language.value, preview: true })); + try { + const { outputs } = await job.result; + const audio = outputs['2']?.audio?.[0]; + if (!audio || scope.signal.aborted) return; + const query = new URLSearchParams({ filename: audio.filename, subfolder: audio.subfolder || '', type: audio.type || 'temp' }); + player.src = `/view?${query}&t=${Date.now()}`; + player.play().catch(() => {}); + } catch (error) { + if (!error.cancelled && !scope.signal.aborted) await anomalousAlert(t('scriptRunFailed', { error: error.message })); + } finally { + job = null; + btn.textContent = label; + btn.classList.remove('is-busy'); + } + }; + + const rowsBox = el('div', 'anomalous-tts-rows'); + const addRow = (from = '', to = '') => { + const line = el('div', 'anomalous-tts-row'); + const fromInput = textInput('anomalous-tts-name-input', from, t('ttsPronounceFromPlaceholder')); + const toInput = textInput('anomalous-tts-text-input', to, t('ttsPronounceToPlaceholder')); + const nowBtn = button('anomalous-tts-listen', t('ttsPronounceNow'), () => preview(fromInput.value, nowBtn), t('ttsPronounceNowHint')); + const afterBtn = button('anomalous-tts-listen', t('ttsPronounceAfter'), () => preview(toInput.value, afterBtn), t('ttsPronounceAfterHint')); + line.append(fromInput, el('span', 'anomalous-tts-arrow', '→'), toInput, nowBtn, afterBtn, + button('anomalous-tts-remove', '×', () => line.remove(), t('ttsEditorRemove'))); + rowsBox.appendChild(line); + return fromInput; + }; + const fillRows = () => { + rowsBox.replaceChildren(); + for (const [from, to] of Object.entries(settings.replace || {})) addRow(from, to); + if (!rowsBox.children.length) addRow(); + }; + fillRows(); + const addBtn = button('anomalous-tts-add', t('ttsPronounceAdd'), () => addRow().focus()); + + const footer = el('div', 'anomalous-voice-modal-footer'); + const saveBtn = button('anomalous-voice-modal-submit', t('ttsEditorSave'), async () => { + if (!loaded) return; + const rows = [...rowsBox.querySelectorAll('.anomalous-tts-row')].map(line => { + const [fromInput, toInput] = line.querySelectorAll('input'); + return { from: fromInput.value, to: toInput.value.trim() }; + }).filter(row => row.from.trim() || row.to); + const problem = validatePronunciationRows(rows); + if (problem) { + await anomalousAlert(problem); + return; + } + const next = { ...settings, format: 1 }; + if (rows.length) next.replace = Object.fromEntries(rows.map(row => [row.from, row.to])); + else delete next.replace; + saveBtn.disabled = true; + try { + const updated = await saveGptSovitsSettings(name, next); + onSaved?.(updated); + close(); + } catch (e) { + if (!scope.signal.aborted) await anomalousAlert(t('ttsEditorSaveFailed', { error: e.message })); + } finally { + if (!scope.signal.aborted) saveBtn.disabled = !loaded; + } + }); + footer.append(button('anomalous-voice-modal-cancel', t('dialogCancel'), close), saveBtn); + + // The summary list may be up to a minute old: read the file as it is now before editing it. + const status = el('div', 'anomalous-tts-hint anomalous-tts-status', t('ttsEditorLoadingFiles')); + saveBtn.disabled = true; + fetchGptSovitsCharacter(name, scope.signal).then(detail => { + if (scope.signal.aborted) return; + if (detail.settings_error) throw new Error(detail.settings_error); + if (detail.settings && typeof detail.settings === 'object') settings = detail.settings; + fillRows(); + status.hidden = true; + loaded = true; + saveBtn.disabled = false; + }).catch(e => { + if (scope.signal.aborted) return; + status.textContent = t('ttsEditorLoadFailed', { error: e.message || String(e) }); + status.classList.add('is-error'); + }); + + const body = el('div', 'anomalous-tts-editor-body'); + body.append(status, el('div', 'anomalous-tts-hint', t('ttsPronounceTips')), languageBar, rowsBox, addBtn); + modal.append(header, body, footer); + overlay.appendChild(modal); + document.body.appendChild(overlay); + rowsBox.querySelector('input')?.focus(); + return close; +} diff --git a/web/styles.css b/web/styles.css index fc816c8..86a3869 100644 --- a/web/styles.css +++ b/web/styles.css @@ -9,7 +9,7 @@ @import url("./styles/08-theme-gallery-overrides.css?v=20260929-audio-tabs"); @import url("./styles/09-prompt-drawers.css?v=20260918-model-sources-1"); @import url("./styles/10-model-sources.css?v=20260918-model-sources-1"); -@import url("./styles/11-audio-studio.css?v=20260926-audio-sidebar-1"); +@import url("./styles/11-audio-studio.css?v=20260929-pronunciation"); @import url("./styles/12-tts-setup.css?v=20260926-tts-import-batch-2"); @import url("./styles/13-script-director.css?v=20260929-generate-roomier"); @import url("./styles/14-version-manager.css"); diff --git a/web/styles/11-audio-studio.css b/web/styles/11-audio-studio.css index 1533ef6..84d9652 100644 --- a/web/styles/11-audio-studio.css +++ b/web/styles/11-audio-studio.css @@ -1571,9 +1571,45 @@ display: none; } +/* Pronunciation table (ui_tts_pronunciation.js): from → to, two listen buttons, remove */ +.anomalous-tts-pronounce .anomalous-tts-row { + grid-template-columns: minmax(0, 1fr) auto minmax(0, 1fr) auto auto auto; +} + +.anomalous-tts-arrow { + color: #64748b; + font-size: 12px; +} + +.anomalous-tts-listen { + border: 1px solid rgba(255, 255, 255, 0.1); + border-radius: 6px; + background: rgba(255, 255, 255, 0.04); + color: #cbd5e1; + cursor: pointer; + font-size: 11px; + padding: 4px 8px; + min-width: 40px; +} + +.anomalous-tts-listen:hover, +.anomalous-tts-listen.is-busy { + border-color: rgba(56, 189, 248, 0.4); + color: #38bdf8; +} + +.anomalous-tts-pronounce-language { + display: flex; + align-items: center; + gap: 8px; + font-size: 12px; + color: #cbd5e1; +} + @media (max-width: 560px) { .anomalous-tts-row, - .anomalous-tts-row.is-main { + .anomalous-tts-row.is-main, + .anomalous-tts-pronounce .anomalous-tts-row { grid-template-columns: 1fr auto; } }