Script for text to speech with text split by token size.

tts_by_token.py

import logging
import os
from pathlib import Path
import sys

import openai
import pygame
from semantic_text_splitter import TextSplitter


# Usage : 
# % open https://platform.openai.com/api-keys
# % export OPENAI_API_KEY="sk-xxxxxx"
# % pbpaste | python tss_by_token.py
openai.api_key = os.getenv("OPENAI_API_KEY")

SPEECH_FILE_PATH = Path(__file__).parent / "output.mp3"

chunk_size = 100
read_speed = 1


def text_to_speech(text: str):
    """Converting text to speech"""
    client = openai.Client()
    # Convert text to speech and get results
    response = client.audio.speech.create(
        model="tts-1",
        voice="alloy",
        speed=read_speed,
        input=text,
    )
    response.stream_to_file(SPEECH_FILE_PATH)


def play_mp3(file_path: Path):
    """Play mp3 files"""
    pygame.mixer.init()
    pygame.mixer.music.load(file_path)
    pygame.mixer.music.play()
    while pygame.mixer.music.get_busy():
        pygame.time.Clock().tick(10)


def text_splitter(text: str, token_size: int) -> list:
    """Split text by the specified number of token size"""
    splitter = TextSplitter.from_tiktoken_model("gpt-3.5-turbo")
    chunks = splitter.chunks(text, token_size)
    return chunks


if __name__ == "__main__":
    text = sys.stdin.read()
    logging.info(f"Input text: len={len(text)}")
    chunks = text_splitter(text, chunk_size)
    for chunk in chunks:
        logging.info(f"chunk: len={len(chunk)}, chunklen={len(chunks)}")
        text_to_speech(chunk)
        play_mp3(SPEECH_FILE_PATH)

You need to install the following dependent libraries before using this script.

[tool.poetry.dependencies]
openai = "^1.13.3"
pygame = "^2.5.2"
semantic-text-splitter = "^0.7.0"