pyminispeaker

Appearance

On this page

Chat Completions ​

This example will show how to play text to speech via the Chat Completions API and pyminispeaker.

Instructions ​

  1. Install a Chat Completions client

We recommend using openai:

sh
pip install openai
  1. In your script, call and obtain any function that allows you to loop over raw audio data. We will be using the following code snippet from OpenAI's Audio API documentation:
python
from openai import OpenAI

API_KEY = "YOUR_API_KEY"
client = OpenAI(api_key=API_KEY)
with client.audio.speech.with_streaming_response.create(
    model="gpt-4o-mini-tts",
    voice="coral",
    input="Today is a wonderful day to build something people love!",
    instructions="Speak in a cheerful and positive tone."
) as response:
    audio_stream = response.iter_bytes(chunk_size=256)
  1. Identify and match the sample rate and sample format of the audio stream.

Since we know their PCM response_format has a sample rate of 24000 and SIGNED16, we can assign these like so:

python
from miniaudio import SampleFormat
SAMPLE_RATE = 24000
SAMPLE_FORMAT = SampleFormat.SIGNED16

WARNING

Selecting the incorrect sample rate and format may result in non-working audio playback.

  1. Pass the audio stream into Speaker with SAMPLE_RATE and SAMPLE_FORMAT.
python
from openai import OpenAI
from minispeaker import Speakers
from miniaudio import SampleFormat

SAMPLE_RATE = 24000
SAMPLE_FORMAT = SampleFormat.SIGNED16

API_KEY = "YOUR_API_KEY"
client = OpenAI(api_key=API_KEY)
speaker = Speakers(sample_rate=SAMPLE_RATE, sample_format=SAMPLE_FORMAT)
with client.audio.speech.with_streaming_response.create(
    model="gpt-4o-mini-tts",
    voice="coral",
    input="Today is a wonderful day to build something people love!",
    response_format="pcm", # <--- NOTE: `pcm` response format is selected to ensure a matching sample rate and sample format. 
    instructions="Speak in a cheerful and positive tone."
) as response:
    audio_stream = response.iter_bytes(chunk_size=256)
    speaker.play(audio_stream, name="My first TTS request") # Play the TTS response
    speaker["My first TTS request"].wait() # Wait for the audio to finish playing

Asynchronous Version ​

python
from openai import AsyncOpenAI
from minispeaker import Speakers
from miniaudio import SampleFormat
import asyncio

SAMPLE_RATE = 24000
SAMPLE_FORMAT = SampleFormat.SIGNED16

API_KEY = "YOUR_API_KEY"
client = AsyncOpenAI(api_key=API_KEY)

async def TTS():
    speaker = Speakers(sample_rate=SAMPLE_RATE, sample_format=SAMPLE_FORMAT)
    async with client.audio.speech.with_streaming_response.create(
        model="gpt-4o-mini-tts",
        voice="coral",
        input="Today is a wonderful day to build something people love!",
        response_format="pcm",
        instructions="Speak in a cheerful and positive tone."
    ) as response:
        audio_stream = response.iter_bytes(chunk_size=256)
        speaker.play(audio_stream, name="My first TTS request") # Play the TTS response
        await speaker["My first TTS request"].wait() # Wait for the audio to finish playing

asyncio.run(TTS())