Gemini tts audio got cutoff

Google Cloud Build with AI Custom ML & MLOps
1

I am developing an app to create a video. The audio part is generated using gemini-2.5-pro-tts using the google genai python sdk. I have this problem where sometimes the generated audio gets cutoff at the last word. For example, the complete phrase is “welcome to my home” but the generated audio stopped mid-word to “welcome to my ho-”. This is happening maybe once in every 10 api call.

Below is minimal code to reproduce the issue

# script.py
import argparse
import asyncio
import os
import wave

from google import genai
from google.genai import types
from google.genai.types import HttpOptions


def save_wave(path: str, audio_data: bytes) -> None:
    os.makedirs(os.path.dirname(path), exist_ok=True)
    with wave.open(path, "wb") as wf:
        wf.setnchannels(1)
        wf.setsampwidth(2)  # 16-bit
        wf.setframerate(24000)
        wf.writeframes(audio_data)


async def generate_tts(text: str, output_path: str, voice_name: str, model: str) -> None:
    client = genai.Client(
        http_options=HttpOptions(api_version="v1"),
        vertexai=True,
        project=os.environ["GOOGLE_CLOUD_PROJECT"],
        location=os.environ.get("GOOGLE_CLOUD_LOCATION", "us-central1"),
    )

    prompt = f"""Please deliver this in a professional, straightforward, friendly, and engaging tone.
Ensure every word is articulated clearly. Include occasional pauses for breath.

Text to synthesize:
{text}"""

    response = await client.aio.models.generate_content(
        model=model,
        contents=prompt,
        config=types.GenerateContentConfig(
            response_modalities=["AUDIO"],
            speech_config=types.SpeechConfig(
                voice_config=types.VoiceConfig(
                    prebuilt_voice_config=types.PrebuiltVoiceConfig(
                        voice_name=voice_name,
                    )
                )
            ),
        ),
    )

    parts = response.candidates[0].content.parts
    audio_data = parts[0].inline_data.data
    save_wave(output_path, audio_data)


async def main() -> None:
    parser = argparse.ArgumentParser()
    parser.add_argument("--prompts-file", required=True)
    parser.add_argument("--output-dir", default="tts_repro_out")
    parser.add_argument("--model", default="gemini-2.0-flash-exp")
    parser.add_argument("--voice", default="Achernar")
    args = parser.parse_args()

    with open(args.prompts_file, "r", encoding="utf-8") as f:
        prompts = [line.strip() for line in f if line.strip()]

    for idx, text in enumerate(prompts):
        out_path = os.path.join(args.output_dir, f"segment_{idx:03d}.wav")
        await generate_tts(text=text, output_path=out_path, voice_name=args.voice, model=args.model)
        print(f"ok index={idx} path={out_path}")


if __name__ == "__main__":
    asyncio.run(main())

run the script with:

python script.py \
  --prompts-file list_of_prompts.txt \
  --output-dir tts_repro_out \
  --model gemini-2.5-pro-tts \
  --voice Achernar

Could anyone help me? Is it wrong code implementation on my side or there is issue from google side?