Audio pipeline dependencies are reduced, bug fixed, room acoustics made more tunable and demo improved #57
Conversation
…e jupyter notebook.
…on pipeline in the jupyter notebook
- Clean pavel fuction header. - Add datasets as a dependency
|
Bug found: Do not merge |
Codecov Report❌ Patch coverage is Additional details and impacted files@@ Coverage Diff @@
## main #57 +/- ##
==========================================
+ Coverage 46.82% 55.19% +8.36%
==========================================
Files 20 34 +14
Lines 4171 5963 +1792
==========================================
+ Hits 1953 3291 +1338
- Misses 2218 2672 +454
🚀 New features to boost your workflow:
|
…inference - Add dsiplay function in AudioDialog - Make __str__ for RecordingDevice - Add some comments to the tutorials - Run all tutorials back - Add local impulse response database in the the tutorial 16 - Add tests data for IR
requirements.txt
Outdated
| # Used to interface with local or hosted large-language-model backends | ||
| ollama | ||
| openai | ||
|
|
There was a problem hiding this comment.
Better if we move these guys to its own requirements-audio.txt so that is then easy to have pip install sdialog[audio]
There was a problem hiding this comment.
I added the requirements-audio.txt and now we need to add the support for pip install sdialog[audio] and add it to the github pipeline.
- Fix weird error in dialog2flow during merging - Update to_audio docs, comment and add an error requiring user to install the audio submodule dependencies. - Update all the relative paths in the tutorials - Fix tutorial 7 not using the downloaded archive - Update .gitignore with new paths - Build the changelog locally
…to-speech" feature - Resample the audio out of the tts when not at self.sampling_rate - Pass extra arguments for tts during inference - Rename the "voice" parameter from the tts generators into "speaker_voice" in order to be unique and prevent clashes with kwargs. - Add comments and simplify the code of the save utterances audios function - Update the documentation - Run the tests - Handle import errors for each BaseTTS subclasses - Remove speed argument from KokoroTTS inference and put it in the constructor - Make the sampling rate from the turns public and saved into the serialized file - Add an alternative for HF TTS in the audio tutorial 1 - Remove useless code in from_turn - Test serialization in both saving and loading for the AudioDialog
.gitignore
Outdated
| #audio files | ||
| *.wav | ||
| *.mp3 | ||
| *.png |
There was a problem hiding this comment.
We do have pngs to be tracked, like images in the docs and in the README, a wouldn't ignore them. If you have pngs somewhere, may be better to ignore them inside that specific folder, but not globally.
There was a problem hiding this comment.
Now we have a subfolder for the audio tutorials, I can simply stop tracking the images in it.
| tutorials/audio/demo_dialog_doctor_patient.json | ||
| tutorials/audio/demo_dialog_doctor_patient_no_age_no_gender.json | ||
| tutorials/audio/customer_support_dialogue.json | ||
| tutorials/audio/=0.9.4 |
There was a problem hiding this comment.
is this one ok? tutorials/audio/=0.9.4
There was a problem hiding this comment.
It's caused by this part of the tutorials using kokoro:
pip install -q kokoro>=0.9.4
apt-get -qq -y install espeak-ng > /dev/null 2>&1
.gitignore
Outdated
| tutorials/audio/dscaper_data_impulse_response/* | ||
| *audio_dialog.json | ||
| ./src/sdialog/sdialog | ||
| tutorials/9.generating_data.ipynb |
There was a problem hiding this comment.
Not sure if it is a good idea to add a tutorial that doesn't exist as part of the global gitignore of the library
There was a problem hiding this comment.
It's a tutorial I have locally only for now
2 participants
No description provided.