runonweb 0.0.1

package/src/tts/supertonic.ts

@@ -0,0 +1,126 @@
+import type { Device, ProgressCallback, ResolvedDevice } from '../core/index.ts'
+import { resolveDevice, toProgressInfo } from '../core/index.ts'
+import { splitUtterances } from './split.ts'
+import { toFloat32 } from './wav.ts'
+import type { SpeakChunk } from './types.ts'
+
+/**
+ * Supertonic 2 (Supertone). One 44.1 kHz voice model for English, Korean, Spanish,
+ * Portuguese and French; 10 preset voices shared across languages. OpenRAIL-M license.
+ */
+const DEFAULT_MODEL = 'onnx-community/Supertonic-TTS-2-ONNX'
+const SAMPLE_RATE = 44_100
+const VOICE_CACHE = 'runonweb-supertonic-voices'
+
+export const SUPERTONIC_LANGUAGES = ['en', 'ko', 'es', 'pt', 'fr'] as const
+export type SupertonicLanguage = (typeof SUPERTONIC_LANGUAGES)[number]
+
+type RawAudio = { audio?: Float32Array | number[]; data?: Float32Array | number[]; sampling_rate?: number }
+
+type SupertonicPipe = {
+  (
+    text: string,
+    options: { speaker_embeddings: Float32Array; num_inference_steps?: number; speed?: number }
+  ): Promise<RawAudio>
+  dispose?: () => Promise<void>
+}
+
+export type LoadedSupertonic = {
+  pipe: SupertonicPipe
+  modelId: string
+  device: ResolvedDevice
+}
+
+export async function loadSupertonic(options: {
+  model?: string
+  device?: Device
+  onProgress?: ProgressCallback
+}): Promise<LoadedSupertonic> {
+  const device = await resolveDevice(options.device ?? 'auto')
+  options.onProgress?.({ status: 'loading', progress: 0 })
+
+  const { pipeline, env } = await import('@huggingface/transformers')
+  env.allowLocalModels = false
+  const modelId = options.model ?? DEFAULT_MODEL
+
+  const pipe = await pipeline('text-to-speech', modelId, {
+    device,
+    dtype: 'fp32',
+    progress_callback: (data: Record<string, unknown>) => {
+      options.onProgress?.(toProgressInfo(data))
+    },
+  })
+
+  options.onProgress?.({ status: 'ready', progress: 100 })
+  return { pipe: pipe as unknown as SupertonicPipe, modelId, device }
+}
+
+export async function* streamSupertonic(
+  st: LoadedSupertonic,
+  text: string,
+  voice: string,
+  speed: number,
+  language: string,
+  steps = 5
+): AsyncGenerator<SpeakChunk> {
+  const lang = (SUPERTONIC_LANGUAGES as readonly string[]).includes(language) ? language : 'en'
+  const style = await loadVoice(st.modelId, voice)
+  for (const piece of splitUtterances(text)) {
+    const raw = await st.pipe(`<${lang}>${piece}</${lang}>`, {
+      speaker_embeddings: style,
+      num_inference_steps: steps,
+      speed,
+    })
+    yield { audio: toFloat32(raw.audio ?? raw.data ?? []), samplingRate: raw.sampling_rate || SAMPLE_RATE, text: piece }
+  }
+}
+
+const voiceCache = new Map<string, Promise<Float32Array>>()
+
+/** Supertonic voice ids are `F1`…`F5` and `M1`…`M5`; we expose them as `st_f1`… */
+export function supertonicVoiceFile(voice: string): string {
+  const m = /^(?:st_)?([fm])([1-5])$/i.exec(voice)
+  if (!m) throw new Error(`Unknown Supertonic voice "${voice}"`)
+  return `${m[1]!.toUpperCase()}${m[2]}`
+}
+
+function loadVoice(modelId: string, voice: string): Promise<Float32Array> {
+  const file = supertonicVoiceFile(voice)
+  const key = `${modelId}/${file}`
+  let pending = voiceCache.get(key)
+  if (!pending) {
+    pending = fetchVoice(modelId, file).catch((err) => {
+      voiceCache.delete(key)
+      throw err
+    })
+    voiceCache.set(key, pending)
+  }
+  return pending
+}
+
+async function fetchVoice(modelId: string, file: string): Promise<Float32Array> {
+  const url = `https://huggingface.co/${modelId}/resolve/main/voices/${file}.bin`
+  let cache: Cache | null = null
+  try {
+    cache = await caches.open(VOICE_CACHE)
+    const hit = await cache.match(url)
+    if (hit) return new Float32Array(await hit.arrayBuffer())
+  } catch {
+    cache = null
+  }
+  const res = await fetch(url)
+  if (!res.ok) throw new Error(`Voice "${file}" not found (${res.status})`)
+  const buffer = await res.arrayBuffer()
+  try {
+    await cache?.put(url, new Response(buffer.slice(0), { headers: { 'content-type': 'application/octet-stream' } }))
+  } catch {
+    // Cache is best-effort.
+  }
+  return new Float32Array(buffer)
+}
+
+export function disposeSupertonic(st: LoadedSupertonic | null) {
+  void st?.pipe.dispose?.()
+}
+
+export { DEFAULT_MODEL as SUPERTONIC_MODEL, SAMPLE_RATE as SUPERTONIC_SAMPLE_RATE }

package/src/tts/types.ts

@@ -0,0 +1,11 @@
+export type SpeakChunk = {
+  audio: Float32Array
+  samplingRate: number
+  text: string
+}
+
+export type TTSResult = {
+  /** Mono PCM samples. 24 kHz for Kokoro / KittenTTS, 44.1 kHz for Supertonic. */
+  audio: Float32Array
+  samplingRate: number
+}

package/src/tts/voices.ts

@@ -0,0 +1,131 @@
+import { DEFAULT_TTS_SIZE, type TTSSize } from './sizes.ts'
+
+export type TTSLocale = 'en-US' | 'en-GB' | 'es' | 'fr' | 'multi'
+export type TTSGender = 'female' | 'male'
+
+export type TTSVoice = {
+  id: string
+  name: string
+  locale: TTSLocale
+  gender: TTSGender
+  /** Subjective grade. A is best. Kitten voices are ungraded. */
+  grade: string
+  size: TTSSize
+}
+
+export const DEFAULT_VOICE = 'af_heart'
+export const DEFAULT_TINY_VOICE = 'bella'
+export const DEFAULT_MULTI_VOICE = 'st_f1'
+
+/** Kokoro v1.0 voices (small) + KittenTTS voices (tiny) + Supertonic 2 presets (multi, any of its 5 languages). */
+export const TTS_VOICES: readonly TTSVoice[] = [
+  { id: 'af_heart', name: 'Heart', locale: 'en-US', gender: 'female', grade: 'A', size: 'small' },
+  { id: 'af_bella', name: 'Bella', locale: 'en-US', gender: 'female', grade: 'A-', size: 'small' },
+  { id: 'af_nicole', name: 'Nicole', locale: 'en-US', gender: 'female', grade: 'B-', size: 'small' },
+  { id: 'af_aoede', name: 'Aoede', locale: 'en-US', gender: 'female', grade: 'C+', size: 'small' },
+  { id: 'af_kore', name: 'Kore', locale: 'en-US', gender: 'female', grade: 'C+', size: 'small' },
+  { id: 'af_sarah', name: 'Sarah', locale: 'en-US', gender: 'female', grade: 'C+', size: 'small' },
+  { id: 'am_fenrir', name: 'Fenrir', locale: 'en-US', gender: 'male', grade: 'C+', size: 'small' },
+  { id: 'am_michael', name: 'Michael', locale: 'en-US', gender: 'male', grade: 'C+', size: 'small' },
+  { id: 'am_puck', name: 'Puck', locale: 'en-US', gender: 'male', grade: 'C+', size: 'small' },
+  { id: 'af_alloy', name: 'Alloy', locale: 'en-US', gender: 'female', grade: 'C', size: 'small' },
+  { id: 'af_nova', name: 'Nova', locale: 'en-US', gender: 'female', grade: 'C', size: 'small' },
+  { id: 'af_sky', name: 'Sky', locale: 'en-US', gender: 'female', grade: 'C-', size: 'small' },
+  { id: 'af_jessica', name: 'Jessica', locale: 'en-US', gender: 'female', grade: 'D', size: 'small' },
+  { id: 'af_river', name: 'River', locale: 'en-US', gender: 'female', grade: 'D', size: 'small' },
+  { id: 'am_echo', name: 'Echo', locale: 'en-US', gender: 'male', grade: 'D', size: 'small' },
+  { id: 'am_eric', name: 'Eric', locale: 'en-US', gender: 'male', grade: 'D', size: 'small' },
+  { id: 'am_liam', name: 'Liam', locale: 'en-US', gender: 'male', grade: 'D', size: 'small' },
+  { id: 'am_onyx', name: 'Onyx', locale: 'en-US', gender: 'male', grade: 'D', size: 'small' },
+  { id: 'am_santa', name: 'Santa', locale: 'en-US', gender: 'male', grade: 'D-', size: 'small' },
+  { id: 'am_adam', name: 'Adam', locale: 'en-US', gender: 'male', grade: 'F+', size: 'small' },
+  { id: 'bf_emma', name: 'Emma', locale: 'en-GB', gender: 'female', grade: 'B-', size: 'small' },
+  { id: 'bf_isabella', name: 'Isabella', locale: 'en-GB', gender: 'female', grade: 'C', size: 'small' },
+  { id: 'bm_george', name: 'George', locale: 'en-GB', gender: 'male', grade: 'C', size: 'small' },
+  { id: 'bm_fable', name: 'Fable', locale: 'en-GB', gender: 'male', grade: 'C', size: 'small' },
+  { id: 'bm_lewis', name: 'Lewis', locale: 'en-GB', gender: 'male', grade: 'D+', size: 'small' },
+  { id: 'bf_alice', name: 'Alice', locale: 'en-GB', gender: 'female', grade: 'D', size: 'small' },
+  { id: 'bf_lily', name: 'Lily', locale: 'en-GB', gender: 'female', grade: 'D', size: 'small' },
+  { id: 'bm_daniel', name: 'Daniel', locale: 'en-GB', gender: 'male', grade: 'D', size: 'small' },
+  { id: 'ef_dora', name: 'Dora', locale: 'es', gender: 'female', grade: 'D', size: 'small' },
+  { id: 'em_alex', name: 'Alex', locale: 'es', gender: 'male', grade: 'D', size: 'small' },
+  { id: 'em_santa', name: 'Santa', locale: 'es', gender: 'male', grade: 'D', size: 'small' },
+  { id: 'ff_siwis', name: 'Siwis', locale: 'fr', gender: 'female', grade: 'B-', size: 'small' },
+  { id: 'bella', name: 'Bella', locale: 'en-US', gender: 'female', grade: '', size: 'tiny' },
+  { id: 'luna', name: 'Luna', locale: 'en-US', gender: 'female', grade: '', size: 'tiny' },
+  { id: 'rosie', name: 'Rosie', locale: 'en-US', gender: 'female', grade: '', size: 'tiny' },
+  { id: 'kiki', name: 'Kiki', locale: 'en-US', gender: 'female', grade: '', size: 'tiny' },
+  { id: 'jasper', name: 'Jasper', locale: 'en-US', gender: 'male', grade: '', size: 'tiny' },
+  { id: 'bruno', name: 'Bruno', locale: 'en-US', gender: 'male', grade: '', size: 'tiny' },
+  { id: 'hugo', name: 'Hugo', locale: 'en-US', gender: 'male', grade: '', size: 'tiny' },
+  { id: 'leo', name: 'Leo', locale: 'en-US', gender: 'male', grade: '', size: 'tiny' },
+  { id: 'st_f1', name: 'F1', locale: 'multi', gender: 'female', grade: '', size: 'multi' },
+  { id: 'st_f2', name: 'F2', locale: 'multi', gender: 'female', grade: '', size: 'multi' },
+  { id: 'st_f3', name: 'F3', locale: 'multi', gender: 'female', grade: '', size: 'multi' },
+  { id: 'st_f4', name: 'F4', locale: 'multi', gender: 'female', grade: '', size: 'multi' },
+  { id: 'st_f5', name: 'F5', locale: 'multi', gender: 'female', grade: '', size: 'multi' },
+  { id: 'st_m1', name: 'M1', locale: 'multi', gender: 'male', grade: '', size: 'multi' },
+  { id: 'st_m2', name: 'M2', locale: 'multi', gender: 'male', grade: '', size: 'multi' },
+  { id: 'st_m3', name: 'M3', locale: 'multi', gender: 'male', grade: '', size: 'multi' },
+  { id: 'st_m4', name: 'M4', locale: 'multi', gender: 'male', grade: '', size: 'multi' },
+  { id: 'st_m5', name: 'M5', locale: 'multi', gender: 'male', grade: '', size: 'multi' },
+]
+
+/** Languages Supertonic 2 (`multi`) can speak with any of its voices. */
+export const TTS_LANGUAGES = {
+  en: 'English',
+  ko: 'Korean',
+  es: 'Spanish',
+  pt: 'Portuguese',
+  fr: 'French',
+} as const
+
+export type TTSLanguage = keyof typeof TTS_LANGUAGES
+
+const LOCALE_LABEL: Record<TTSLocale, string> = {
+  'en-US': 'American English',
+  'en-GB': 'British English',
+  es: 'Spanish',
+  fr: 'French',
+  multi: 'Supertonic · English, Korean, Spanish, Portuguese, French',
+}
+
+export function voicesFor(size: TTSSize = DEFAULT_TTS_SIZE): TTSVoice[] {
+  return TTS_VOICES.filter((v) => v.size === size)
+}
+
+export function defaultVoiceFor(size: TTSSize = DEFAULT_TTS_SIZE): string {
+  return size === 'tiny' ? DEFAULT_TINY_VOICE : size === 'multi' ? DEFAULT_MULTI_VOICE : DEFAULT_VOICE
+}
+
+export function getVoice(id: string): TTSVoice | undefined {
+  return TTS_VOICES.find((v) => v.id === id)
+}
+
+export const TTS_VOICE_GROUPS = voiceGroupsFor(DEFAULT_TTS_SIZE)
+
+export function voiceGroupsFor(size: TTSSize = DEFAULT_TTS_SIZE) {
+  const voices = voicesFor(size)
+  const locales = [...new Set(voices.map((v) => v.locale))]
+  return locales.map((locale) => ({
+    locale,
+    label: LOCALE_LABEL[locale],
+    voices: voices.filter((v) => v.locale === locale),
+  }))
+}
+
+export function voiceLabel(voice: TTSVoice): string {
+  if (voice.locale === 'multi') return `${voice.name} · ${voice.gender}`
+  const region =
+    voice.locale === 'en-GB' ? 'British' : voice.locale === 'es' ? 'Spanish' : voice.locale === 'fr' ? 'French' : 'American'
+  const grade = voice.grade ? `${voice.gender} · ${voice.grade}` : voice.gender
+  return `${voice.name} · ${region} ${grade}`
+}
+
+/** eSpeak language for Kokoro voices that `phonemizer` cannot handle (Spanish, French). */
+export function kokoroPhonemeLang(voiceId: string): 'es' | 'fr-fr' | null {
+  const prefix = voiceId.slice(0, 2)
+  if (prefix === 'ef' || prefix === 'em') return 'es'
+  if (prefix === 'ff') return 'fr-fr'
+  return null
+}

package/src/tts/wav.ts

@@ -0,0 +1,52 @@
+/** Encode mono Float32 PCM as a 16-bit WAV Blob. */
+export function float32ToWavBlob(samples: Float32Array, sampleRate: number): Blob {
+  const numChannels = 1
+  const bytesPerSample = 2
+  const blockAlign = numChannels * bytesPerSample
+  const buffer = new ArrayBuffer(44 + samples.length * bytesPerSample)
+  const view = new DataView(buffer)
+
+  writeString(view, 0, 'RIFF')
+  view.setUint32(4, 36 + samples.length * bytesPerSample, true)
+  writeString(view, 8, 'WAVE')
+  writeString(view, 12, 'fmt ')
+  view.setUint32(16, 16, true)
+  view.setUint16(20, 1, true)
+  view.setUint16(22, numChannels, true)
+  view.setUint32(24, sampleRate, true)
+  view.setUint32(28, sampleRate * blockAlign, true)
+  view.setUint16(32, blockAlign, true)
+  view.setUint16(34, 8 * bytesPerSample, true)
+  writeString(view, 36, 'data')
+  view.setUint32(40, samples.length * bytesPerSample, true)
+
+  let offset = 44
+  for (let i = 0; i < samples.length; i++) {
+    const s = Math.max(-1, Math.min(1, samples[i]!))
+    view.setInt16(offset, s < 0 ? s * 0x8000 : s * 0x7fff, true)
+    offset += 2
+  }
+
+  return new Blob([buffer], { type: 'audio/wav' })
+}
+
+export function concatFloat32(chunks: Float32Array[]): Float32Array {
+  const total = chunks.reduce((n, c) => n + c.length, 0)
+  const out = new Float32Array(total)
+  let offset = 0
+  for (const chunk of chunks) {
+    out.set(chunk, offset)
+    offset += chunk.length
+  }
+  return out
+}
+
+export function toFloat32(data: Float32Array | number[] | ArrayLike<number>): Float32Array {
+  return data instanceof Float32Array ? data : new Float32Array(data)
+}
+
+function writeString(view: DataView, offset: number, str: string) {
+  for (let i = 0; i < str.length; i++) {
+    view.setUint8(offset + i, str.charCodeAt(i))
+  }
+}