开发者社区 > ModelScope模型即服务 > 语音 > 正文

我调用 tts 模型推理,声音结尾感觉像被阉割一样短促的结束,而同样的文字放在句子中间则没事,为啥?

其中有报错如下:

Load pinyin_en_mix_dict failed
Load pinyin_en_mix_dict failed
Load pinyin_en_mix_dict failed
Load pinyin_en_mix_dict failed
Load pinyin_en_mix_dict failed
Load pinyin_en_mix_dict failed
Load pinyin_en_mix_dict failed
Load pinyin_en_mix_dict failed
Load pinyin_en_mix_dict failed
Load pinyin_en_mix_dict failed
Load pinyin_en_mix_dict failed
text.cc: festival_Text_init
2023-08-22 17:29:40,024 - modelscope - WARNING - No preprocessor field found in cfg.
[2023-08-22 17:29:40] [WARNING] [base.py:283] No preprocessor field found in cfg.
2023-08-22 17:29:40,024 - modelscope - WARNING - No val key and type key found in preprocessor domain of configuration.json file.
[2023-08-22 17:29:40] [WARNING] [base.py:292] No val key and type key found in preprocessor domain of configuration.json file.
2023-08-22 17:29:40,024 - modelscope - WARNING - Cannot find available config to build preprocessor at mode inference, current config: {'model_dir': '/home/grow/models/modelscope/hub/damo/speech_sambert-hifigan_tts_zhitian_emo_zh-cn_16k'}. trying to build by task and model information.
[2023-08-22 17:29:40] [WARNING] [base.py:311] Cannot find available config to build preprocessor at mode inference, current config: {'model_dir': '/home/grow/models/modelscope/hub/damo/speech_sambert-hifigan_tts_zhitian_emo_zh-cn_16k'}. trying to build by task and model information.
2023-08-22 17:29:40,024 - modelscope - WARNING - No preprocessor key ('sambert-hifigan', 'text-to-speech') found in PREPROCESSOR_MAP, skip building preprocessor.
[2023-08-22 17:29:40] [WARNING] [base.py:325] No preprocessor key ('sambert-hifigan', 'text-to-speech') found in PREPROCESSOR_MAP, skip building preprocessor.
FormatProsodicUtterance failed
FormatProsodicUtterance failed
FormatProsodicUtterance failed
Removing weight norm...
[W NNPACK.cpp:64] Could not initialize NNPACK! Reason: Unsupported hardware.

展开
收起
1229914862415005 2023-08-22 17:31:02 1035 1
3 条回答
写回答
取消 提交回答
  • 我也面临同样的问题,这个有什么解决方案吗?

    2024-05-15 13:50:23
    赞同 展开评论 打赏
  • 请问可以离线的部署吗

    2024-04-22 18:48:55
    赞同 展开评论 打赏
  • 北京阿里云ACE会长

    我已经认真阅读了 你的问题:

    【 我调用 tts 模型推理,声音结尾感觉像被阉割一样短促的结束,而同样的文字放在句子中间则没事,为啥?

    并思考了

    建议如下:


    可能是因为在进行 TTS 模型推理时,某些字符或词汇的读音受到了影响,导致声音结尾短促。对于报错的问题,可能是由于加载拼音词典失败导致的。建议您检查一下拼音词典文件是否存在,或者尝试重新安装拼音词典。

    2023-08-24 08:08:47
    赞同 展开评论 打赏

包括语音识别、语音合成、语音唤醒、声学设计及信号处理、声纹识别、音频事件检测等多个领域

相关电子书

更多
大规模机器学习在蚂蚁+阿里的应用 立即下载
阿里巴巴机器学习平台AI 立即下载
微博机器学习平台架构和实践 立即下载