Anvia

Appearance

Audio generation ​

Audio generation turns text into audio bytes through an AudioGenerationModel. Use it for narration, accessibility, spoken summaries, and media previews.

Create the provider model as shown in Audio generation models, then pass it into the server-side workflow that owns voice and format policy.

Generate speech ​

ts
import { audioGenerationRequest } from '@anvia/core/audio-generation'

const response = await audioGenerationRequest(audioModel)
  .text('Your incident summary is ready for review.')
  .voice('alloy')
  .speed(1)
  .additionalParams({ response_format: 'mp3' })
  .send()

The normalized response contains audio as a Uint8Array, an optional mediaType, and the raw provider response. Set a voice explicitly in production; the core builder cannot choose a product-appropriate voice for you.

Provider-specific formats and synthesis settings belong in additionalParams(...). Validate those settings against an allow-list before they reach the request builder.

Control generated scripts ​

Do not send unbounded model output directly into speech generation. Validate the final script after any completion or agent stage:

ts
import { z } from 'zod'

const SpeechRequest = z.object({
  text: z.string().min(1).max(3_000),
  voice: z.enum(['alloy', 'verse', 'aria']),
})

const input = SpeechRequest.parse(requestedSpeech)

const speech = await audioGenerationRequest(audioModel)
  .text(input.text)
  .voice(input.voice)
  .speed(1)
  .additionalParams({ response_format: 'mp3' })
  .send()

The allow-list is application policy, not a universal provider list. Use only voices supported by the configured model and approved for your product.

Store and serve the result ​

Persist the audio in object storage or a media service, then return an asset ID or short-lived URL:

ts
const asset = await mediaStore.put({
  bytes: speech.audio,
  mediaType: speech.mediaType ?? 'audio/mpeg',
  metadata: {
    voice: input.voice,
    model: audioModel.defaultModel ?? 'unknown',
  },
})

Keep raw audio out of sessions, traces, and event logs. Use a durable worker for long scripts, bulk generation, or workflows that need progress and retries. When a product imitates or represents a real person, add the required consent, disclosure, and review controls outside the model request.

Built for Anvia.