Comments (2)
If you are interested, I opened a PR yesterday (#8723) which contains a tutorial notebook you could download and try https://github.com/NVIDIA/NeMo/blob/79ffef97d2993b716406227f00fc234c19541610/tutorials/tts/Audio_Codec_Training.ipynb. Change the branch from 'main' to 'codec_tutorial'.
The training manifest only needs {"audio_filepath", "duration"}
The metadata structure is defined at https://github.com/NVIDIA/NeMo/blob/main/nemo/collections/tts/data/vocoder_dataset.py#L45. For each dataset you use it asks for the manifest_path
to the manifest file, and the root audio_dir
where the audio is stored. If training on multiple datasets you can specify sample_weight
to determine how often it samples from each.
train_ds_meta
configures training data, val_ds_meta
configures data for computing validation statistics/loss, and log_ds_meta
is used with this callback (https://github.com/NVIDIA/NeMo/blob/main/nemo/collections/tts/parts/utils/callbacks.py#L331) to configure logging of various artifacts during training (eg. reconstructed audio, embedding plots). If doing local logging then log_dir
specifies the local directory.
A full training command (copied from the tutorial), with 1 dataset "vctk", looks like:
python /content/nemo/examples/tts/audio_codec.py
--config-path=/content/nemo/examples/tts/conf/audio_codec
--config-name=audio_codec_24000.yaml
max_epochs=10
weighted_sampling_steps_per_epoch=10
batch_size=4
log_dir=/content/exps/EnCodec/test_run/logs
exp_manager.exp_dir=/content/exps
+exp_manager.version=test_run
model.log_config.log_wandb=False
model.log_config.log_tensorboard=False
model.log_config.generators.0.log_dequantized=True
trainer.accelerator=gpu
+train_ds_meta.vctk.manifest_path=/content/data/vctk_subset_multispeaker/train_manifest.json
+train_ds_meta.vctk.audio_dir=/content/data/vctk_subset_multispeaker/audio_preprocessed
+val_ds_meta.vctk.manifest_path=/content/data/vctk_subset_multispeaker/dev_manifest.json
+val_ds_meta.vctk.audio_dir=/content/data/vctk_subset_multispeaker/audio_preprocessed
+log_ds_meta.vctk.manifest_path=/content/data/vctk_subset_multispeaker/dev_manifest.json
+log_ds_meta.vctk.audio_dir=/content/data/vctk_subset_multispeaker/audio_preprocessed
Let me know if you have any other questions.
from nemo.
Thank you for your kind reply. It was so helpful to me !!
from nemo.
Related Issues (20)
- Runtime error when exporting FastPitch model to PyTorch
- support for ndarray and tensor decoding for RNNT based ASR models HOT 10
- Unable to train with more than 0 dataloader workers HOT 2
- Issue running PEFT finetuning on a GPT2 model with TP=2 and SP =True HOT 2
- NeMo ASR - Model level config does not contain `sample_rate`, please explicitly provide `sample_rate` to the dataloaders. HOT 2
- Not able to run basic inference using canary-1b model HOT 8
- numba.cuda.cudadrv.error.NvvmError: Failed to verify
- ModuleNotFoundError: No module named 'nemo.collections.asr.metrics.rnnt_wer' HOT 7
- hybrid_rnnt_ctc_bpe model has no inference code HOT 1
- gpt SFT model inference did not return the prediction results
- No module named 'nemo' HOT 2
- Exporting Conformer CTC to TensorRT HOT 3
- CONF-TSASR HOT 1
- Tutorials broken due to Megatron-Core dependency HOT 2
- Support for multi-channel audio data HOT 3
- ckpt to nemo or model HOT 2
- Restoring from pretrained checkpoint is broke for "stt_ua_fastconformer_hybrid_large_pc" HOT 4
- parakeet-tdt-1.1b unable to calculate ASR confidence scores! HOT 2
- assertion error on calling model_eval.load_adapters() in lora tutorial HOT 2
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from nemo.