microsoft/semantic-kernel · error · ServiceResponseException

{type(self)} service failed to generate audio

Error message

{type(self)} service failed to generate audio

What it means

Raised as ServiceResponseException in _send_text_to_audio_request when any exception occurs during client.audio.speech.create. This is the text-to-speech (TTS) path; failures include invalid model/voice parameters, text exceeding limits, content-filter rejections, or network/quota errors.

Source

Thrown at python/semantic_kernel/connectors/ai/open_ai/services/open_ai_handler.py:190

        except Exception as ex:
            raise ServiceResponseException(
                f"{type(self)} service failed to transcribe audio",
                ex,
            ) from ex

    async def _send_text_to_audio_request(
        self, settings: OpenAITextToAudioExecutionSettings
    ) -> _legacy_response.HttpxBinaryResponseContent:
        """Send a request to the OpenAI text to audio endpoint.

        The OpenAI API returns the content of the generated audio file.
        """
        try:
            return await self.client.audio.speech.create(
                **settings.prepare_settings_dict(),
            )
        except Exception as ex:
            raise ServiceResponseException(
                f"{type(self)} service failed to generate audio",
                ex,
            ) from ex

    def _handle_structured_output(
        self, request_settings: OpenAIChatPromptExecutionSettings, settings: dict[str, Any]
    ) -> None:
        response_format = getattr(request_settings, "response_format", None)
        if getattr(request_settings, "structured_json_response", False) and response_format:
            # Case 1: response_format is a type and subclass of BaseModel
            if isinstance(response_format, type) and issubclass(response_format, BaseModel):
                settings["response_format"] = type_to_response_format_param(response_format)
            # Case 2: response_format is a type but not a subclass of BaseModel
            elif isinstance(response_format, type):
                generated_schema = KernelJsonSchemaBuilder.build(parameter_type=response_format, structured_output=True)
                assert generated_schema is not None  # nosec
                settings["response_format"] = generate_structured_output_response_format_schema(
                    name=response_format.__name__, schema=generated_schema

View on GitHub (pinned to c028a0c7dc)

Solutions

1. Inspect ex.__cause__ for the specific API error
2. Verify the model id is 'tts-1' or 'tts-1-hd' and the voice is one of the supported names
3. Ensure input text is within the 4096-character limit
4. For content-filter rejections, sanitize or shorten the input text

Example fix

# before
settings = OpenAITextToAudioExecutionSettings(ai_model_id='tts-1', voice='custom_voice')
# after — use a supported voice
settings = OpenAITextToAudioExecutionSettings(ai_model_id='tts-1', voice='alloy')

SUPPORTED_VOICES = {'alloy', 'echo', 'fable', 'onyx', 'nova', 'shimmer'}
SUPPORTED_TTS_MODELS = {'tts-1', 'tts-1-hd'}
if settings.ai_model_id not in SUPPORTED_TTS_MODELS:
    raise ValueError(f'Unsupported TTS model: {settings.ai_model_id}')
if settings.voice and settings.voice not in SUPPORTED_VOICES:
    raise ValueError(f'Unsupported voice: {settings.voice}')
if len(settings.input_text) > 4096:
    raise ValueError('TTS input exceeds 4096 character limit')

from semantic_kernel.exceptions import ServiceResponseException

try:
    audio = await service._send_text_to_audio_request(settings)
except ServiceResponseException as e:
    logger.error('TTS generation failed: %s', e)
    raise

When it happens

Trigger: Calling text-to-speech with unsupported voice names, invalid model id (e.g., not 'tts-1' or 'tts-1-hd'), text that triggers content filtering, response_format not supported, or network/rate-limit failures during client.audio.speech.create.

Common situations: Using a voice name not in OpenAI's supported set (alloy, echo, fable, onyx, nova, shimmer); very long input text exceeding the 4096-character limit; content filter on the TTS input; rate limit from high-frequency speech generation.

Related errors

• The voice '{voice}' is not supported.
• The format '{format}' is not supported.
• {type(self)} service failed to generate embeddings
• Failed to generate image: {ex}
• Failed to edit image: {ex}

AI-assisted analysis of microsoft/semantic-kernel@c028a0c7dc (2026-08-13). Data as JSON: /api/errors/ca659a6523ea43c7. Report an issue: GitHub.