[Bug] TypeError: expected str, bytes or os.PathLike object, not NoneType

[vitaliy-sharandin]

Describe the bug

I was validating the resolution of bug #3224 which I suspect was fixed in commit 4d0f53d and got a new unexpected error from the same code. I empirically figured that error appears after changes in commit preceding the one that fixed my issue (8c5227e), hence I can't really validate the fix.

Error log:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
[<ipython-input-8-37c7de7f3028>](https://localhost:8080/#) in <cell line: 2>()
      1 tts = TTS('tts_models/multilingual/multi-dataset/xtts_v2', gpu=True)
----> 2 tts.tts_with_vc(text=f"Как оно?", speaker_wav='/content/SPEAKER_00_voice_clips.wav', language='ru')

13 frames
[/usr/local/lib/python3.10/dist-packages/TTS/api.py](https://localhost:8080/#) in tts_with_vc(self, text, language, speaker_wav, speaker)
    464         with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as fp:
    465             # Lazy code... save it to a temp file to resample it while reading it for VC
--> 466             self.tts_to_file(text=text, speaker=speaker, language=language, file_path=fp.name)
    467         if self.voice_converter is None:
    468             self.load_vc_model_by_name("voice_conversion_models/multilingual/vctk/freevc24")

[/usr/local/lib/python3.10/dist-packages/TTS/api.py](https://localhost:8080/#) in tts_to_file(self, text, speaker, language, speaker_wav, emotion, speed, pipe_out, file_path, **kwargs)
    401                 pipe_out=pipe_out,
    402             )
--> 403         wav = self.tts(text=text, speaker=speaker, language=language, speaker_wav=speaker_wav, **kwargs)
    404         self.synthesizer.save_wav(wav=wav, path=file_path, pipe_out=pipe_out)
    405         return file_path

[/usr/local/lib/python3.10/dist-packages/TTS/api.py](https://localhost:8080/#) in tts(self, text, speaker, language, speaker_wav, emotion, speed, **kwargs)
    339                 text=text, speaker_name=speaker, language=language, emotion=emotion, speed=speed
    340             )
--> 341         wav = self.synthesizer.tts(
    342             text=text,
    343             speaker_name=speaker,

[/usr/local/lib/python3.10/dist-packages/TTS/utils/synthesizer.py](https://localhost:8080/#) in tts(self, text, speaker_name, language_name, speaker_wav, style_wav, style_text, reference_wav, reference_speaker_name, **kwargs)
    376             for sen in sens:
    377                 if hasattr(self.tts_model, "synthesize"):
--> 378                     outputs = self.tts_model.synthesize(
    379                         text=sen,
    380                         config=self.tts_config,

[/usr/local/lib/python3.10/dist-packages/TTS/tts/models/xtts.py](https://localhost:8080/#) in synthesize(self, text, config, speaker_wav, language, **kwargs)
    390 
    391         """
--> 392         return self.inference_with_config(text, config, ref_audio_path=speaker_wav, language=language, **kwargs)
    393 
    394     def inference_with_config(self, text, config, ref_audio_path, language, **kwargs):

[/usr/local/lib/python3.10/dist-packages/TTS/tts/models/xtts.py](https://localhost:8080/#) in inference_with_config(self, text, config, ref_audio_path, language, **kwargs)
    412         }
    413         settings.update(kwargs)  # allow overriding of preset settings with kwargs
--> 414         return self.full_inference(text, ref_audio_path, language, **settings)
    415 
    416     @torch.inference_mode()

[/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py](https://localhost:8080/#) in decorate_context(*args, **kwargs)
    113     def decorate_context(*args, **kwargs):
    114         with ctx_factory():
--> 115             return func(*args, **kwargs)
    116 
    117     return decorate_context

[/usr/local/lib/python3.10/dist-packages/TTS/tts/models/xtts.py](https://localhost:8080/#) in full_inference(self, text, ref_audio_path, language, temperature, length_penalty, repetition_penalty, top_k, top_p, do_sample, gpt_cond_len, gpt_cond_chunk_len, max_ref_len, sound_norm_refs, **hf_generate_kwargs)
    473             Sample rate is 24kHz.
    474         """
--> 475         (gpt_cond_latent, speaker_embedding) = self.get_conditioning_latents(
    476             audio_path=ref_audio_path,
    477             gpt_cond_len=gpt_cond_len,

[/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py](https://localhost:8080/#) in decorate_context(*args, **kwargs)
    113     def decorate_context(*args, **kwargs):
    114         with ctx_factory():
--> 115             return func(*args, **kwargs)
    116 
    117     return decorate_context

[/usr/local/lib/python3.10/dist-packages/TTS/tts/models/xtts.py](https://localhost:8080/#) in get_conditioning_latents(self, audio_path, max_ref_length, gpt_cond_len, gpt_cond_chunk_len, librosa_trim_db, sound_norm_refs, load_sr)
    349         speaker_embedding = None
    350         for file_path in audio_paths:
--> 351             audio = load_audio(file_path, load_sr)
    352             audio = audio[:, : load_sr * max_ref_length].to(self.device)
    353             if sound_norm_refs:

[/usr/local/lib/python3.10/dist-packages/TTS/tts/models/xtts.py](https://localhost:8080/#) in load_audio(audiopath, sampling_rate)
     70 
     71     # torchaudio should chose proper backend to load audio depending on platform
---> 72     audio, lsr = torchaudio.load(audiopath)
     73 
     74     # stereo to mono if needed

[/usr/local/lib/python3.10/dist-packages/torchaudio/_backend/utils.py](https://localhost:8080/#) in load(uri, frame_offset, num_frames, normalize, channels_first, format, buffer_size, backend)
    201         """
    202         backend = dispatcher(uri, format, backend)
--> 203         return backend.load(uri, frame_offset, num_frames, normalize, channels_first, format, buffer_size)
    204 
    205     return load

[/usr/local/lib/python3.10/dist-packages/torchaudio/_backend/ffmpeg.py](https://localhost:8080/#) in load(uri, frame_offset, num_frames, normalize, channels_first, format, buffer_size)
    332             )
    333         else:
--> 334             return load_audio(os.path.normpath(uri), frame_offset, num_frames, normalize, channels_first, format)
    335 
    336     @staticmethod

[/usr/lib/python3.10/posixpath.py](https://localhost:8080/#) in normpath(path)
    338 def normpath(path):
    339     """Normalize path, eliminating double slashes, etc."""
--> 340     path = os.fspath(path)
    341     if isinstance(path, bytes):
    342         sep = b'/'

TypeError: expected str, bytes or os.PathLike object, not NoneType

To Reproduce

Code:

tts = TTS('tts_models/multilingual/multi-dataset/xtts_v2', gpu=True)
tts.tts_with_vc_to_file(text=f"Как оно?", speaker_wav='/content/drive/MyDrive/Data/SPEAKER_00_voice_clips.wav', language='ru', file_path=f'/content/drive/MyDrive/Data/test.wav')

OR

tts = TTS('tts_models/multilingual/multi-dataset/xtts_v2', gpu=True)
tts.tts_with_vc(text=f"Как оно?", speaker_wav='/content/drive/MyDrive/Data/SPEAKER_00_voice_clips.wav', language='ru')

Environment

- TTS from commit 8c5227ed8489ba1ae528371a6df46de77a144333 and onward
- Google Colab environment

[eginhard]

Just use the following with XTTS. It has voice cloning integrated and the output doesn't need to be passed through another voice conversion model (#3293 tracks providing a better error message in this case).

tts = TTS('tts_models/multilingual/multi-dataset/xtts_v2', gpu=True)
tts.tts_to_file(text=f"Как оно?", speaker_wav='/content/drive/MyDrive/Data/SPEAKER_00_voice_clips.wav', language='ru', file_path=f'/content/drive/MyDrive/Data/test.wav')

# OR

tts = TTS('tts_models/multilingual/multi-dataset/xtts_v2', gpu=True)
tts.tts(text=f"Как оно?", speaker_wav='/content/drive/MyDrive/Data/SPEAKER_00_voice_clips.wav', language='ru')

[vitaliy-sharandin]

True, README is also up to date, still hard to notice method names updates among all those changes. Thank you.