Vercel AI SDK

1
import { experimental_generateSpeech as generateSpeech } from 'ai';
2
import { smallestai } from 'smallestai-vercel-provider';
3
4
const { audio } = await generateSpeech({
5
  model: smallestai.speech('lightning-v3.1'),
6
  text: 'Hello from Smallest AI!',
7
  voice: 'sophia',
8
  speed: 1.0,
9
});
10
11
// audio.uint8Array — raw WAV bytes
12
// audio.base64 — base64 encoded audio

1
import { experimental_transcribe as transcribe } from 'ai';
2
import { smallestai } from 'smallestai-vercel-provider';
3
import { readFileSync } from 'fs';
4
5
const { text, segments } = await transcribe({
6
  model: smallestai.transcription('pulse'),
7
  audio: readFileSync('recording.wav'),
8
  mediaType: 'audio/wav',
9
});
10
11
console.log(text);       // "Hello from Smallest AI!"
12
console.log(segments);   // [{ text: "Hello", startSecond: 0, endSecond: 0.5 }, ...]

1
// app/api/speak/route.ts
2
import { experimental_generateSpeech as generateSpeech } from 'ai';
3
import { smallestai } from 'smallestai-vercel-provider';
4
5
export async function POST(req: Request) {
6
  const { text, voice } = await req.json();
7
8
  const { audio } = await generateSpeech({
9
    model: smallestai.speech('lightning-v3.1'),
10
    text,
11
    voice: voice || 'sophia',
12
  });
13
14
  return new Response(Buffer.from(audio.uint8Array), {
15
    headers: { 'Content-Type': 'audio/wav' },
16
  });
17
}

1
const res = await fetch('/api/speak', {
2
  method: 'POST',
3
  body: JSON.stringify({ text: 'Hello!', voice: 'sophia' }),
4
});
5
const blob = await res.blob();
6
new Audio(URL.createObjectURL(blob)).play();

1
const { audio } = await generateSpeech({
2
  model: smallestai.speech('lightning-v3.1'),
3
  text: 'Hello!',
4
  voice: 'robert',
5
  providerOptions: {
6
    smallestai: {
7
      sampleRate: 48000,         // 8000 | 16000 | 24000 | 44100 | 48000
8
      outputFormat: 'mp3',       // pcm | mp3 | wav | mulaw
9
      language: 'en',            // ISO 639-1 code
10
      consistency: 0.5,          // voice consistency (0–1)
11
      similarity: 0.5,           // voice similarity (0–1)
12
      enhancement: 1,            // audio enhancement level (0–2)
13
    },
14
  },
15
});

1
const result = await transcribe({
2
  model: smallestai.transcription('pulse'),
3
  audio: audioBuffer,
4
  mediaType: 'audio/wav',
5
  providerOptions: {
6
    smallestai: {
7
      language: 'hi',            // ISO 639-1 code
8
      diarize: true,             // speaker identification
9
      emotionDetection: true,    // detect emotions
10
      ageDetection: true,        // detect speaker age
11
      genderDetection: true,     // detect speaker gender
12
    },
13
  },
14
});