播报机器人----阅读失败 #288
Description
问题描述
版本：super-agent-party-win-v0.3.1
操作：机器人---->播报机器人--->粘贴测试文本：你好--->点击开始生成
观察现象：进度条正常，vrm字幕显示正常，但是语音没有合成
TTS策略：
1.本地使用index-tts1.5，正常对话会生成语音，但是这里播报没有
2.发现接口没有被调用
希望解决：
能否给出对应修改位置
接口内容如下：
`
@app.get("/tts")
async def synthesize(
text: str = "欢迎使用语音合成接口",
infer_mode: str = "普通推理",
max_text_tokens_per_sentence: int = 120,
sentences_bucket_max_size: int = 4,
do_sample: bool = True,
top_p: float = 0.8,
top_k: int = 30,
temperature: float = 1.0,
length_penalty: float = 0.0,
num_beams: int = 3,
repetition_penalty: float = 10.0,
max_mel_tokens: int = 600
):
"""单次推理接口"""
try:
global COUNT
# 创建输出目录
os.makedirs("outputs", exist_ok=True)
# # 保存上传的参考音频
# prompt_path = f"temp_prompt_{int(time.time())}.wav"
# with open(prompt_path, "wb") as f:
# f.write(await prompt_audio.read())
#
# 使用本地固定的参考音频
prompt_path = r"G:\AAA_myIdea\indexTTS\index-tts\zh.wav"
if not os.path.exists(prompt_path):
return {"error": f"参考音频不存在：{prompt_path}"}
# 准备输出路径 循环1000次覆盖
COUNT += 1
if COUNT > 1000:
COUNT = 0
output_path = os.path.join("outputs", f"api_output_{COUNT}.wav")
# 调用生成逻辑
kwargs = {
"do_sample": do_sample,
"top_p": top_p,
"top_k": top_k if top_k > 0 else None,
"temperature": temperature,
"length_penalty": length_penalty,
"num_beams": num_beams,
"repetition_penalty": repetition_penalty,
"max_mel_tokens": max_mel_tokens,
}
if infer_mode == "普通推理":
tts.infer(
prompt_path,
text,
output_path,
max_text_tokens_per_sentence=max_text_tokens_per_sentence,
**kwargs
)
else:
tts.infer_fast(
prompt_path,
text,
output_path,
max_text_tokens_per_sentence=max_text_tokens_per_sentence,
sentences_bucket_max_size=sentences_bucket_max_size,
**kwargs
)
# os.remove(prompt_path)
# return FileResponse(output_path, media_type="audio/wav")
# 返回文件，background参数会在响应完成后删除文件（可选）
return FileResponse(
output_path,
media_type="audio/wav",
headers={
"Content-Disposition": "inline", # 改为 inline 让浏览器直接播放
"Cache-Control": "no-cache"
}
)
except Exception as e:
raise HTTPException(status_code=500, detail=str(e))`