Documentation Index
Fetch the complete documentation index at: https://docs.getpatter.com/llms.txt
Use this file to discover all available pages before exploring further.
ElevenLabs TTS (REST)
ElevenLabsTTS is the default ElevenLabs TTS adapter. It streams audio chunks over the HTTP /v1/text-to-speech/{voiceId}/stream endpoint and exposes the same synthesize / synthesizeStream shape as the rest of Patter’s TTS providers.
REST vs WebSocket — when to use which
ElevenLabs ships two streaming transports. Patter wraps both:
| Transport | Class | Use when |
|---|
| HTTP / REST | ElevenLabsTTS (this page) | Default. Simpler, more forgiving on flaky networks, supports every model including eleven_v3*. |
| WebSocket | ElevenLabsWebSocketTTS | You’re latency-bound on TTS time-to-first-byte and willing to give up eleven_v3 support. Saves ~50 ms per utterance vs REST. |
eleven_v3. The REST adapter is the canonical choice and gets every new ElevenLabs model on day one.
Install
ElevenLabsTTS ships in the core getpatter package — no extras needed:
Quickstart
// npx tsx example.ts
import { Patter, Twilio, DeepgramSTT, ElevenLabsTTS } from "getpatter";
const phone = new Patter({ carrier: new Twilio(), phoneNumber: "+15550001234" });
const agent = phone.agent({
stt: new DeepgramSTT(), // DEEPGRAM_API_KEY from env
tts: new ElevenLabsTTS({ voiceId: "rachel" }), // ELEVENLABS_API_KEY from env
systemPrompt: "You are a helpful assistant.",
});
await phone.serve({ agent });
Carrier-tuned factories
For real phone calls always reach for the carrier-specific factories — they pre-set the right outputFormat so the SDK can skip resampling and PCM ↔ μ-law transcoding on the audio path:
import { ElevenLabsTTS } from "getpatter";
// Twilio: ulaw_8000 + speaker_boost off, moderate stability for clean μ-law
const ttsTwilio = ElevenLabsTTS.forTwilio({ voiceId: "rachel" });
// Telnyx: pcm_16000 (matches L16/16000 default Telnyx codec)
const ttsTelnyx = ElevenLabsTTS.forTelnyx({ voiceId: "rachel" });
forTwilio saves ~30–80 ms first-byte plus per-frame CPU and removes a potential aliasing source. If your Telnyx profile is pinned to PCMU/8000 instead, construct directly with outputFormat: "ulaw_8000".
Constructor
new ElevenLabsTTS({
apiKey?: string; // ELEVENLABS_API_KEY from env when omitted
voiceId?: string; // default "21m00Tcm4TlvDq8ikWAM" (Rachel)
modelId?: ElevenLabsModel | string; // default "eleven_flash_v2_5"
outputFormat?: ElevenLabsOutputFormat; // default "pcm_16000"
voiceSettings?: ElevenLabsVoiceSettings;
languageCode?: string;
chunkSize?: number; // default 4096
})
interface ElevenLabsVoiceSettings {
stability?: number;
similarity_boost?: number;
style?: number;
use_speaker_boost?: boolean;
}
| Parameter | Default | Description |
|---|
apiKey | env | ElevenLabs API key. |
voiceId | "21m00Tcm4TlvDq8ikWAM" (Rachel) | Voice ID or curated display name ("rachel", "alloy", …). |
modelId | "eleven_flash_v2_5" | See Models. |
outputFormat | "pcm_16000" | See Output formats. |
voiceSettings | — | { stability, similarity_boost, style, use_speaker_boost }. |
languageCode | — | BCP-47 / ISO 639-1 language hint. |
chunkSize | 4096 | Stream chunk size in bytes. |
API surface
// Buffer the full utterance
const audio: Buffer = await tts.synthesize("Hello!");
// Or stream chunks
for await (const chunk of tts.synthesizeStream("Hello!")) {
forwardToCaller(chunk);
}
synthesizeStream cancels the underlying HTTP stream when the consumer breaks out of the loop, so ElevenLabs stops billing for unconsumed characters.
Models
| Model ID | Notes |
|---|
eleven_flash_v2_5 (default) | Fastest TTFT (~75 ms). Best for live phone calls. |
eleven_turbo_v2_5 | Balanced quality / speed. |
eleven_multilingual_v2 | Best multilingual support. |
eleven_monolingual_v1 | Legacy English-only. |
eleven_v3 | Newest, highest quality. Slower TTFT. Not supported by the WebSocket variant. |
as const object for autocomplete:
import { ElevenLabsModel } from "getpatter";
ElevenLabsModel.FLASH_V2_5 // "eleven_flash_v2_5"
ElevenLabsModel.V3 // "eleven_v3"
| Format | Use case |
|---|
pcm_16000 (default) | Web playback, dashboard previews, Telnyx L16/16000 carrier path. |
ulaw_8000 | Twilio Media Streams (μ-law @ 8 kHz native). |
pcm_8000 / pcm_22050 / pcm_24000 / pcm_44100 | Other PCM rates. |
mp3_22050_32 … mp3_44100_192 | MP3 at various bitrates. |
import { ElevenLabsOutputFormat } from "getpatter";
ElevenLabsOutputFormat.ULAW_8000 // "ulaw_8000"
ElevenLabsOutputFormat.PCM_16000 // "pcm_16000"
Voice ID resolution
Pass either an opaque 20-char ElevenLabs voice ID ("21m00Tcm4TlvDq8ikWAM") or a curated display name. The SDK ships a built-in name → ID map for common voices ("rachel", "bella", "matilda", etc.) plus an "alloy" alias for cross-provider portability with OpenAI naming. Unknown strings are passed through unchanged so custom voices keep working.
Pricing
| Model | $/1k chars |
|---|
eleven_flash_v2_5 (default) | 0.06 |
eleven_turbo_v2_5 | 0.05 |
eleven_multilingual_v2 | 0.18 |
eleven_monolingual_v1 | 0.18 |
eleven_v3 | 0.30 |
DEFAULT_PRICING per model — no override needed unless you’re on a custom ElevenLabs plan. See Metrics for how to register custom rates.
See also
