@spekoai/ai-sdk-provider

Speko provider for the Vercel AI SDK - routed speech and transcription across every voice provider.

Speko is an OpenRouter-style routing layer for the voice stack (TTS + STT). One API key, one provider entry in your AI SDK code, and Speko's benchmark-driven router picks the best upstream provider (ElevenLabs, Cartesia, Deepgram, AssemblyAI, OpenAI, and more) per request - and fails over automatically when a provider degrades.

Implements the AI SDK provider specification v4 (SpeechModelV4 + TranscriptionModelV4), for ai v7.

Install

npm install @spekoai/ai-sdk-provider ai

Setup

Get an API key at platform.speko.dev and set it as SPEKO_API_KEY, or pass it explicitly:

import { createSpeko } from '@spekoai/ai-sdk-provider';

const speko = createSpeko({
  apiKey: process.env.SPEKO_API_KEY,
  // baseURL: 'https://api.speko.dev',
});

The default instance reads SPEKO_API_KEY from the environment:

import { speko } from '@spekoai/ai-sdk-provider';

Speech

import { speko } from '@spekoai/ai-sdk-provider';
import { generateSpeech } from 'ai';

const { audio } = await generateSpeech({
  model: speko.speech('auto'),
  text: 'Welcome to Speko.',
  voice: 'QtY3JBOUKEB5xzrRfOKc', // optional, provider-specific
  providerOptions: {
    speko: { language: 'en', optimizeFor: 'latency' },
  },
});

// audio.uint8Array - WAV bytes (Speko synthesizes raw PCM; this provider
// wraps it into a WAV container at the routed provider's sample rate).

Pass outputFormat: 'pcm' to receive the raw s16le mono PCM instead of WAV. The sample rate is reported in providerMetadata.speko.sampleRate.

Transcription

import { speko } from '@spekoai/ai-sdk-provider';
import { transcribe } from 'ai';
import { readFile } from 'node:fs/promises';

const result = await transcribe({
  model: speko.transcription('auto'),
  audio: await readFile('call.wav'),
  providerOptions: {
    speko: {
      language: 'es-MX',
      keywords: ['Speko', 'Vercel'],
    },
  },
});

console.log(result.text);
console.log(result.providerMetadata.speko);
// { provider: 'deepgram', model: 'nova-3', confidence: 0.93, failoverCount: 0, scoresRunId: '...' }

Model ids

The model id selects a routing strategy, not a single model:

| Model id | Behavior | | --- | --- | | auto | Router default for your organization's routing policy | | auto-fast | Bias toward latency | | auto-quality | Bias toward accuracy/quality | | auto-cheap | Bias toward cost | | deepgram | Pin routing and failover to one upstream provider | | elevenlabs/eleven_multilingual_v2 | Pin the provider and request a specific upstream model |

speko.transcription('auto-quality');
speko.speech('elevenlabs/eleven_multilingual_v2');

Routing options (providerOptions.speko)

This is where Speko differs from single-vendor providers: the options control the router, not one model.

| Option | Applies to | Description | | --- | --- | --- | | language | both | BCP-47 language tag, e.g. en, es-MX. Routing is language-aware. Defaults to en | | region | both | Region to rank providers in, e.g. us-east4, eu-west1 | | optimizeFor | both | balanced | accuracy | latency | cost; overrides the model id preset | | providers | both | Restrict routing AND failover to these upstream providers | | keywords | transcription | Domain keywords forwarded to the routed STT provider | | model | speech | Upstream model override, e.g. sonic-2 | | spokenForm | speech | Deterministic spoken-form normalization before synthesis |

Failover

Failover is handled server-side by Speko: if the routed provider errors or degrades, the request is retried against the next-ranked provider before the call returns. The number of failovers taken is reported in providerMetadata.speko.failoverCount. Use providers to control the pool the router (and its failover) may use.

Notes

• The one-shot transcription endpoint does not report per-word timing yet: when the input duration is deterministic (WAV or raw PCM input) the provider reports durationInSeconds and a single whole-transcript segment; otherwise segments is empty and durationInSeconds is undefined.
• languageModel, embeddingModel, and imageModel throw NoSuchModelError - Speko is a voice platform.

Links

• Speko documentation
• Dashboard and API keys
• TypeScript SDK (@spekoai/sdk)

License

MIT