Files
Hermes-ui/packages/server/src/services/hermes/tts.ts
T

79 lines
2.0 KiB
TypeScript
Raw Normal View History

2026-05-08 15:34:11 +08:00
import { EdgeTTS } from 'node-edge-tts'
import { tmpdir } from 'os'
import { join } from 'path'
import { readFile, unlink } from 'fs/promises'
import { randomUUID } from 'crypto'
import { logger } from '../logger'
const FIXED_VOICE = 'zh-CN-XiaoxiaoNeural'
const FIXED_RATE = '+4%'
const FIXED_PITCH = '+12Hz'
export interface TtsOptions {
text: string
lang?: string
voice?: string
rate?: string
pitch?: string
2026-05-08 15:34:11 +08:00
}
export async function edgeTts(opts: TtsOptions): Promise<Buffer> {
const id = randomUUID()
const tmpFile = join(tmpdir(), `tts-${id}.mp3`)
try {
const tts = new EdgeTTS({
voice: opts.voice || FIXED_VOICE,
rate: opts.rate || FIXED_RATE,
pitch: opts.pitch || FIXED_PITCH,
2026-05-08 15:34:11 +08:00
timeout: 15000,
})
await tts.ttsPromise(opts.text, tmpFile)
const buf = await readFile(tmpFile)
return buf
} finally {
unlink(tmpFile).catch(() => {})
}
}
export async function textToSpeech(opts: TtsOptions): Promise<{ audio: Buffer; engine: string }> {
const voice = opts.voice || FIXED_VOICE
const rate = opts.rate || FIXED_RATE
const pitch = opts.pitch || FIXED_PITCH
2026-05-08 15:34:11 +08:00
const audio = await edgeTts(opts)
logger.debug({ engine: 'edge', voice, rate, pitch }, 'TTS generated via Edge')
2026-05-08 15:34:11 +08:00
return { audio, engine: 'edge' }
}
/**
* Convert speed multiplier (0.5-2.0) to Edge TTS rate string.
* Edge TTS rate format: "+/-NN%"
*/
export function speedToEdgeRate(speed: number): string {
const percent = Math.round((speed - 1) * 100)
return percent >= 0 ? `+${percent}%` : `${percent}%`
}
/**
* Convert OpenAI TTS request to internal TtsOptions.
* OpenAI format: { model, input, voice, speed }
*/
export interface OpenaiTtsRequest {
model?: string
input: string
voice?: string
speed?: number
}
export async function openaiCompatibleTts(
body: OpenaiTtsRequest,
): Promise<{ audio: Buffer; engine: string }> {
return textToSpeech({
text: body.input,
voice: body.voice || FIXED_VOICE,
rate: body.speed ? speedToEdgeRate(body.speed) : FIXED_RATE,
pitch: FIXED_PITCH,
})
}