OpenAI TTS

OpenAITTS is a Patter TTSProvider backed by OpenAI’s POST /v1/audio/speech endpoint. It streams 24 kHz PCM from OpenAI and resamples it to 16 kHz or 8 kHz on the fly so the telephony output path can forward bytes without an additional resample stage. No vendor SDK required.

Install

openai-tts ships in the base install — no extra needed.

pip install getpatter

On Python 3.13+ install audioop-lts for the in-process 24 kHz → 16 kHz (or 8 kHz) resampler. It’s pinned in getpatter’s base dependencies.

Usage

Use the namespaced import (getpatter.tts.openai) or the flat alias (getpatter.OpenAITTS). Both auto-resolve OPENAI_API_KEY from the environment when api_key= is omitted.

# Namespaced import
from getpatter.tts import openai

tts = openai.TTS()                                        # reads OPENAI_API_KEY
tts = openai.TTS(api_key="sk-...", voice="nova", model="tts-1-hd")

# Flat alias (equivalent)
from getpatter import OpenAITTS

tts = OpenAITTS()

// See typescript-sdk/providers/openai-tts for the TS twin.

In an agent:

import asyncio
from getpatter import Patter, Twilio, DeepgramSTT, OpenAITTS

phone = Patter(carrier=Twilio(), phone_number="+15550001234")

agent = phone.agent(
    stt=DeepgramSTT(),                                    # DEEPGRAM_API_KEY from env
    tts=OpenAITTS(voice="nova"),                          # OPENAI_API_KEY from env
    system_prompt="You are a helpful assistant.",
)

asyncio.run(phone.serve(agent))

Models and rates

OpenAI TTS bills per 1,000 characters synthesized. Per-model rates (defaults from getpatter.pricing):

Model	Rate / 1k chars	Notes
`tts-1` (default)	$0.015	Fast, telephony-friendly.
`tts-1-hd`	$0.030	Higher fidelity — heavier.
`gpt-4o-mini-tts`	$0.012	Accepts `instructions` for voice direction.

(gpt-4o-mini-tts is billed by tokens upstream; published per 1k chars equivalent for parity with the rest of the table. Override per-call via Patter(pricing={"openai_tts": {"models": {"tts-1": {"price": ...}}}}).)

Voices

Built-in voices accepted by all three models: alloy (default), ash, ballad, coral, echo, fable, nova, onyx, sage, shimmer, verse.

Languages

OpenAI TTS automatically follows the language of the input text — no language parameter exists. The same voice can synthesize across most of OpenAI’s supported languages (English, Spanish, French, German, Italian, Portuguese, Japanese, Chinese, …) — see the OpenAI voice guide for the full coverage matrix.

Voice direction (`gpt-4o-mini-tts` only)

Pass instructions= to steer tone, pacing, and accent:

tts = OpenAITTS(
    api_key="sk-...",
    model="gpt-4o-mini-tts",
    voice="sage",
    instructions="Speak in a warm, empathetic tone, like a customer support specialist.",
)

tts-1 and tts-1-hd reject instructions with a 400 — the SDK silently drops the field for those models so you can swap freely.

Options

Option	Default	Notes
`api_key`	`None`	Reads from `OPENAI_API_KEY` when omitted.
`model`	`"tts-1"`	`"tts-1"`, `"tts-1-hd"`, `"gpt-4o-mini-tts"`.
`voice`	`"alloy"`	Any of the 11 built-in voices above.
`instructions`	`None`	Voice-direction prompt (ignored for legacy models).
`speed`	`None`	Float in `[0.25, 4.0]`.
`target_sample_rate`	`16000`	`8000` or `16000` Hz — chooses the resampler chain.

Low-level usage

from getpatter.providers.openai_tts import OpenAITTS

tts = OpenAITTS(api_key="sk-...", voice="nova", model="tts-1")
async for chunk in tts.synthesize("Hello from the Patter pipeline."):
    ...
await tts.close()

​OpenAI TTS

​Install

​Usage

​Models and rates

​Voices

​Languages

​Voice direction (gpt-4o-mini-tts only)

​Options

​Low-level usage

OpenAI TTS

Install

Usage

Models and rates

Voices

Languages

Voice direction (`gpt-4o-mini-tts` only)

Options

Low-level usage