Default Podcast Generation Fails Due to Hardcoded Voice Types in tts_node

[ZhiyuXu0124]

Description

It appears that the BV002_streaming and BV001_streaming voice types are no longer available in the Volcengine TTS service. This is confirmed by referencing the official voice type list: Volcengine TTS Voice Types. As a result, the default podcast generation process fails.

Initially, I attempted to update the VOLCENGINE_TTS_VOICE_TYPE in the .env file to resolve the issue. However, the error persisted, as shown in the logs below:

2025-05-11 20:45:27,796 - src.podcast.graph.tts_node - ERROR - {'reqid': 'xxxxxxxxxxxxxxxx', 'code': 3001, 'message': '[resource_id=volc.tts.default] requested resource not granted'}

Upon further investigation, I found that the voice types for different speakers are hardcoded in the src/podcast/graph/tts_node.py file. This is causing the issue since the unavailable voice types are directly referenced in the code.

Relevant Code

Here is the relevant code block from src/podcast/graph/tts_node.py:

def tts_node(state: PodcastState):
    logger.info("Generating audio chunks for podcast...")
    tts_client = _create_tts_client()
    for line in state["script"].lines:
        tts_client.voice_type = (
            "BV002_streaming" if line.speaker == "male" else "BV001_streaming"
        )
        result = tts_client.text_to_speech(line.paragraph, speed_ratio=1.05)
        if result["success"]:
            audio_data = result["audio_data"]
            audio_chunk = base64.b64decode(audio_data)
            state["audio_chunks"].append(audio_chunk)
        else:
            logger.error(result["error"])
    return {
        "audio_chunks": state["audio_chunks"],
    }

Suggestions for Improvement

To address this issue and make the system more flexible, I propose the following changes:

1. Configuration via .env: Allow all voice type configurations to be managed through environment variables. For example:
   • VOLCENGINE_TTS_VOICE_TYPE_MALE
   • VOLCENGINE_TTS_VOICE_TYPE_FEMALE
2. Enhanced Podcast Configuration: Introduce a more dynamic configuration system for podcast generation. This could include:
   • The number of speakers.
   • Role assignments for each speaker.
   • Customizable voice types for each role.

[jizhi0v0]

https://console.volcengine.com/speech/app

在这里编辑你所用的app，并且授予「语音合成大模型」下的「大模型语音合成」权限

[ZhiyuXu0124]

https://console.volcengine.com/speech/app

在这里编辑你所用的app，并且授予「语音合成大模型」下的「大模型语音合成」权限

火山这边的应用我一直在用的，目前测试看下来是火山下架了你们默认的音色，我改用我之前克隆的声音一切就都正常了。主要是发现你们代码中写死了男女生的音色，导致env的配置不起作用了

[hahazei]

配置自己购买的音色类型
VOLCENGINE_TTS_CLUSTER=volcano_tts # Optional, default is volcano_tts
VOLCENGINE_TTS_VOICE_TYPE=BV005_streaming # Optional, default is BV700_V2_streaming