Speech Recognition With CitriNet

Automatic Speech Recognition (ASR) models take in audio files and predict their transcriptions. Besides Jasper and QuartzNet, we can also use CitriNet for ASR. CitriNet is a successor of QuartzNet that features on sub-word tokenization and better backbone architecture.

Downloading Sample Spec Files

Example specification files for the following ASR tasks can be downloaded using the following command:

tao speech_to_text_citrinet download_specs -o <target_path> \
                                   -r <results_path>

Required Arguments

  • -o: Target path where the spec files will be stored.

  • -r: Results and output log directory.

Preparing the Dataset

The dataset for ASR consists of a set of utterances in individual audio files (.wav) and a manifest that describes the dataset, with information about one utterance per line (.json).

Each line of the manifest should be in the following format:

{"audio_filepath": "/path/to/audio.wav", "text": "the transcription of the utterance", "duration": 23.147}

The audio_filepath field should provide an absolute path to the .wav file corresponding to the utterance. The text field should contain the full transcript for the utterance, and the duration field should reflect the duration of the utterance in seconds.

Each entry in the manifest (describing one audio file) should be bordered by ‘{‘ and ‘}’ and must be contained on one line. The fields that describe the file should be separated by commas, and have the form "field_name": value, as shown above.

Since the manifest specifies the path for each utterance, the audio files do not have to be located in the same directory as the manifest, or even in any specific directory structure.

Creating an Experiment Spec File

The spec file for ASR using CitriNet includes the trainer, save_to, model, training_ds, validation_ds, and optim parameters. The following is a shortened example of a spec file for training on the Mozilla Common Voice English dataset.

trainer:
  max_epochs: 100
tlt_checkpoint_interval: 1
# Name of the .tlt file where the trained CitriNet model will be saved
save_to: trained-model.tlt

# Specifies parameters for the ASR model
model:
  # Parameters for sub-word tokenization
  tokenizer:
    dir: ???
    type: "bpe"  # Can be either bpe or wpe

  # Parameters for the audio to spectrogram preprocessor.
  preprocessor:
    _target_: nemo.collections.asr.modules.AudioToMelSpectrogramPreprocessor
    normalize: per_feature
    sample_rate: 16000
    window_size: 0.02
    window_stride: 0.01
    window: hann
    features: 64
    n_fft: 512
    frame_splicing: 1
    dither: 1.0e-05
    stft_conv: false

  # This adds spectrogram augmentation to the training process.
  spec_augment:
    _target_: nemo.collections.asr.modules.SpectrogramAugmentation
    rect_masks: 5
    rect_freq: 50
    rect_time: 120

  # The encoder and decoder sections specify your model architecture.
  encoder:
    _target_: nemo.collections.asr.modules.ConvASREncoder
    feat_in: 64
    activation: relu
    conv_mask: true

    # Several blocks were cut out here for brevity.
    jasper:
    - filters: 128
      repeat: 1
      kernel: [11]
      stride: [1]
      dilation: [1]
      dropout: 0.0
      residual: true
      separable: true
      se: true
      se_context_size: -1

      #... (Add more blocks to describe the model)

      - filters: &enc_feat_out 1024
      repeat: 1
      kernel: [1]
      stride: [1]
      dilation: [1]
      dropout: 0.0
      residual: false
      separable: true
      se: true
      se_context_size: -1


  decoder:
    _target_: nemo.collections.asr.modules.ConvASRDecoder
    feat_in: 1024
    num_classes: -1  # filled with vocabulary size from tokenizer at runtime
    vocabulary: []  # filled with vocabulary from tokenizer at runtime

# This section specifies the dataset to be used for training.
training_ds:
  # No need to specify an audio file path, since the manifest entries contain individual
  # utterances' file paths.
  manifest_filepath: /data/cv-corpus-5.1-2020-06-22/en/train.json
  sample_rate: 16000
  batch_size: 32
  trim_silence: true
  # Setting a max duration trims out files that are longer than the max.
  max_duration: 16.7
  shuffle: true
  # The is_tarred and tarred_audio_filepaths parameters should be specified if using a tarred dataset.
  is_tarred: false
  tarred_audio_filepaths: null

# Specifies the dataset to be used for validation.
validation_ds:
  manifest_filepath: /data/cv-corpus-5.1-2020-06-22/en/dev.json
  sample_rate: 16000
  batch_size: 32
  shuffle: false

# The parameters for the training optimizer, including learning rate, lr schedule, etc.
optim:
  name: adam
  lr: .1
  # optimizer arguments
  betas: [0.9, 0.999]
  weight_decay: 0.0001
  # scheduler setup
  sched:
    name: CosineAnnealing
    # scheduler config override
    warmup_steps: null
    warmup_ratio: 0.05
    min_lr: 1e-6
    last_epoch: -1

The specification can be roughly grouped into three categories:

  • Parameters that describe the training process

  • Parameters that describe the datasets, and

  • Parameters that describe the model.

This specification can be used with the tao speech_to_text_citrinet train command. Only a dataset parameter is required for tao speech_to_text_citrinet evaluate, though a checkpoint must be provided.

If you would like to change a parameter for your run without changing the specification file itself, you can specify it on the command line directly. For example, if you would like to change the validation batch size, you can add validation_ds.batch_size=1 to your command, which would override the batch size of 32 in the configuration shown above. An example of this is shown in the training instructions below.

Training Process Configs

There are a few parameters that help specify the parameters of your training run, which are detailed in the following table.

Parameter

Datatype

Description

Supported Values

max_epochs

int

Specifies the maximum number of epochs to train the model. A field for the trainer parameter.

>0

save_to

string

The location to save the trained model checkpoint. Should be in the form path/to/target/location/modelname.tlt.

Valid paths.

optim

Specifies the optimizer to be used for training, as well as the parameters to configure it, including:

  • name (String): Which optimizer to use.

  • lr (float): The learning rate. Must be specified.

  • sched: Specifies learning rate schedule, if desired.

If your chosen optimizer takes additional arguments, they can be placed under lr, as seen in the example above with betas and weight_decay.

tlt_checkpoint_interval

int

The interval (number of epochs) to save the .tlt checkpoint during training.

>=0 (0 means no checkpoint)

Dataset Configs

The datasets that you use should be specified by <xyz>_ds parameters, depending on your use case:

  • For training using tao speech_to_text_citrinet train, you should have training_ds to describe your training dataset, and validation_ds to describe your validation dataset.

  • For evaluation using tao speech_to_text_citrinet evaluate, you should have test_ds to describe your test dataset.

  • For fine-tuning using tao speech_to_text_citrinet finetune, you should have finetuning_ds to describe the fine-tuning training dataset, and validation_ds to describe your validation dataset.

The fields for each dataset config are described in the following table.

Parameter

Datatype

Description

Supported Values

manifest_filepath

string

The filepath to the manifest (.json file) that describes the audio data.

Valid filepaths.

sample_rate

int

Target sample rate to load the audio, in kHz.

batch_size

int

Batch size. This may depend on memory size and how long your audio samples are.

>0

trim_silence

bool

Whether or not to trim silence from the beginning and end of each audio signal. Defaults to false if no value is set.

True/False

min_duration

float

All files with a duration less than the given value (in seconds) will be dropped. Defaults to 0.1.

max_duration

float

All files with a duration greater than the given value (in seconds) will be dropped.

shuffle

bool

Whether or not to shuffle the data. We recommend true for training data, and false for validation.

True/False

is_tarred

bool

Whether the audio files in the dataset are contained in a tarfile (.tar). If so, you must also set tarred_audio_filepaths, and set shuffle_n if you would like the data to be shuffled. Defaults to false.

True/False

tarred_audio_filepaths

string

Only to be set if is_tarred is set to true. Path to the tarfile (.tar) that contains the audio samples associated with the entries in manifest_filepath.

Valid filepaths.

shuffle_n

int

Only to be set if is_tarred is set to true. The number of audio samples to load at once from the tarfile for shuffling. For example, if set to 100 when batch size is 25, the data loader will load the next 100 samples in the tarfile, shuffle them, and use the shuffled order for the next four batches.

Model Configs

Your CitriNet model architecture and configuration are set under the model parameter. This includes parameters for tokenizer, which defines the tokenizer type and path for sub-word tokenization; parameters for the audio preprocessor, which determines how audio signals are transformed to spectrograms; spectrogram augmentation, which adds a data augmentation step to the pipeline; the encoder of the model; and the decoder of the model.

The tokenizer parameters are as follows:

Parameter

Datatype

Description

Supported Values

dir

string

Root path to the tokneizer model. This path is assumed to be created by the create_tokenizer command.

Valid path.

type

string

Type of the tokenizer, either ‘bpe’ or ‘wpe’.

The preprocessor parameters are as follows:

Parameter

Datatype

Description

Supported Values

normalize

string

How to normalize each spectrogram. Defaults to per_feature.

  • per_feature: Normalizes each spectrogram per channel/frequency.

  • all_features: Normalizes over the entire spectrogram to be mean 0 with std 1.

  • Any other value: Disables normalization.

sample_rate

int

Sample rate of the input audio data in kHz. This should match your datasets’ sample rates. Defaults to 16000.

window_size

float

Window size for FFT in seconds. Defaults to 0.02.

window_stride

float

Window stride for FFT in seconds. Defaults to 0.01.

window

string

Windowing function for FFT. Defaults to hann.

hann, hamming, blackman, bartlett

features

int

Number of mel spectrogram frequency bins to output. Defaults to 64.

n_fft

int

Length of FFT window.

frame_splicing

int

How many frames to stack across the feature dimension. Setting this to 1 disables frame splicing. Defaults to 1.

dither

float

Amount of white-noise dithering. Defaults to 1e-5.

stft_conv

bool

If set to true, uses pytorch_stft and convolutions. If set to false, uses torch.stft. Defaults to false.

If you would like to add spectrogram augmentation to your model, then you can include a spec_augment block. Within this block, you can specify parameters for time and frequency cuts for augmentation, as described by SpecAugment and Cutout.

Parameter

Datatype

Description

Supported Values

rect_masks

int

How many rectangular masks should be cut (Cutout). Defaults to 5.

rect_freq

int

Should only be set if rect_masks was set. Maximum size of cut rectangles along the frequency dimension. Defaults to 5.

rect_time

int

Should only be set if rect_masks was set. Maximum size of cut rectangles along the time dimension. Defaults to 25.

freq_masks

int

How many frequency segments should be cut (SpecAugment). Defaults to 0.

freq_width

int

Should only be set if freq_masks is set. Maximum number of frequencies to be cut in one segment. Defaults to 10.

time_masks

int

How many time segments should be cut (SpecAugment). Defaults to 0.

time_width

int

Should only be set if time_masks is set. Maximum number of time steps to be cut in one segment. Defaults to 10.

The encoder parameters for your model include details about the CitriNet encoder architecture, including how many blocks to use, how many times to repeat each block, and convolution parameters per block.

To use CitriNet (which uses squeeze-and-excitation(SE) blocks), add separable: true, se: true, and se_context_size: -1 to all the blocks in the architecture. (Note: do not change the parameter name jasper.)

The encoder parameters are detailed in the following table.

Parameter

Datatype

Description

Supported Values

feat_in

int

The number of input features. Should be equal to features in the preprocessor parameters.

activation

string

What activation function to use in the encoder.

hardtanh, relu, selu, swish

conv_mask

bool

Whether to use masked convolutions in the encoder. Defaults to false.

jasper

A list of blocks that specifies your encoder architecture. Each entry in this list represents one block in the architecture and contains the parameters for that block, including convolution parameters, dropout, and the number of times the block is repeated.

The decoder parameters are detailed in the following table.

Parameter

Datatype

Description

Supported Values

feat_in

int

The number of input features to the decoder. Should be equal to the number of filters in the last block of the encoder.

vocabulary

list

A list of the valid output characters for your model. For CitriNet, this should always be an empty list.

num_classes

int

Number of output classes. For CitriNet this should always be set to -1.

Subword Tokenization with the Tokenizer

Before we can do the actual training, we need to do some processings to the text. This step is called subword tokenization that creates a subword vocabulary for the text. This is different from Jasper/QuartzNet because only single characters are regarded as elements in the vacabulary in their cases, while in CitriNet the subword can be one or multiple characters. We can use the create_tokenizer command to create the tokenizer that can generate the subword vocabulary for us for use in training below.

As mentioned above, you can add additional arguments to override configurations from your experiment specification file. This allows you to create valid spec files that leave these fields blank, to be specified as command line arguments at runtime.

tao speech_to_text_citrinet create_tokenizer \
    -e <experiment_spec> \
    manifests=<manifest_file_paths> \
    output_root=<output_dir> \
    vocab_size=<vocaburary_size>

Required Arguments

  • -e: The experiment specification file to set up tokenizer, described in detail below.

Creating a config file for tokenizer

The command create_tokenizer requires a config file. This config file is also in yaml format. It contains manifests, output_root, vocab_size, and tokenizer parameters in it, as in table below.

Parameter

Datatype

Description

Supported Values

manifests

string

Comma separated list of manifest file paths, can be one or more. The manifest file should be the same as one used in training.

output_root

string

The output root path for the tokenizer model to be generated.

vocab_size

int

The vocaburary size of the tokenizer.

The tokenizer parameter has its own sub-parameters, as in table below.

Parameter

Datatype

Description

Supported Values

tokenizer_type

string

Type of the tokenizer, currently 'spe' and 'wpe' are supported.

'spe' or 'wpe'

spe_type

string

Sub-type of ‘spe’ tokenizer. Valid only when tokenizer_type is set to 'spe'.

unigram, bpe, char, word

spe_character_coverage

float

How much of the original vocabulary it should cover in its “base set” of tokens.

<=1

lower_case

bool

Create separate tokens for upper and lower case characters or not. Set it to True will not create separated tokens for upper case and lower case characters.

Training the Model

To train a model from scratch, use the following command:

tao speech_to_text_citrinet train -e <experiment_spec> -g <num_gpus>

As mentioned above, you can add additional arguments to override configurations from your experiment specification file. This allows you to create valid spec files that leave these fields blank, to be specified as command line arguments at runtime.

For example, the following command can be used to override the training manifest and validation manifest, the number of epochs to train, and the place to save the model checkpoint:

tao speech_to_text_citrinet train -e <experiment_spec> \
                          -g <num_gpus>  \
                          training_ds.manifest_filepath=<training_manifest_filepath> \
                          validation_ds.manifest_filepath=<val_manifest_filepath> \
                          trainer.max_epochs=<epochs_to_train> \
                          save_to='<file_path>.tlt'

Required Arguments

  • -e: The experiment specification file to set up training, as in the example given above.

Optional Arguments

  • -g: The number of GPUs to be used in the training in a multi-GPU scenario (default: 1).

  • -r: The path to the results and log directory. Log files, checkpoints, etc., will be stored here.

  • -k: The key to encrypt the model.

  • Other arguments to override fields in the specification file.

Training Procedure

At the start of each training experiment, TAO Toolkit will print out a log of the experiment specification, including any parameters added or overridden via the command line. It will also show additional information, such as which GPUs are available, where logs will be saved, how many hours are in each loaded dataset, and how much of each dataset has been filtered.

GPU available: True, used: True
TPU available: None, using: 0 TPU cores
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0,1,2]
[NeMo W 2021-01-20 21:41:53 exp_manager:375] Exp_manager is logging to ./, but it already exists.
[NeMo I 2021-01-20 21:41:53 exp_manager:323] Resuming from checkpoints/trained-model-last.ckpt
[NeMo I 2021-01-20 21:41:53 exp_manager:186] Experiments will be logged at .
...
[NeMo I 2021-01-20 21:41:54 collections:173] Dataset loaded with 948 files totaling 0.71 hours
[NeMo I 2021-01-20 21:41:54 collections:174] 0 files were filtered totaling 0.00 hours
[NeMo I 2021-01-20 21:41:54 collections:173] Dataset loaded with 130 files totaling 0.10 hours
[NeMo I 2021-01-20 21:41:54 collections:174] 0 files were filtered totaling 0.00 hours

You should next see a full printout of the number of parameters in each module and submodule, as well as the total number of trainable and non-trainable parameters in the model.

In the following table, the encoder module contains 18.9 million parameters, and its submodule encoder.encoder.0, the Jasper block, contains 19,000 of those parameters. Of those 19,000 parameters, 2.1k are from the first MaskedConv1d, 16.4k are from the second, and 512 are from BatchNorm1d. Also listed are the ReLU and dropout submodules, with no parameters.

| Name                                 | Type                              | Params
-----------------------------------------------------------------------------------------
0   | preprocessor                     | AudioToMelSpectrogramPreprocessor | 0
1   | preprocessor.featurizer          | FilterbankFeatures                | 0
2   | encoder                          | ConvASREncoder                    | 18.9 M
3   | encoder.encoder                  | Sequential                        | 18.9 M
4   | encoder.encoder.0                | JasperBlock                       | 19.0 K
5   | encoder.encoder.0.mconv          | ModuleList                        | 19.0 K
6   | encoder.encoder.0.mconv.0        | MaskedConv1d                      | 2.1 K
7   | encoder.encoder.0.mconv.0.conv   | Conv1d                            | 2.1 K
8   | encoder.encoder.0.mconv.1        | MaskedConv1d                      | 16.4 K
9   | encoder.encoder.0.mconv.1.conv   | Conv1d                            | 16.4 K
10  | encoder.encoder.0.mconv.2        | BatchNorm1d                       | 512
11  | encoder.encoder.0.mout           | Sequential                        | 0
12  | encoder.encoder.0.mout.0         | ReLU                              | 0
13  | encoder.encoder.0.mout.1         | Dropout                           | 0
...
600 | decoder                          | ConvASRDecoder                    | 29.7 K
601 | decoder.decoder_layers           | Sequential                        | 29.7 K
602 | decoder.decoder_layers.0         | Conv1d                            | 29.7 K
603 | loss                             | CTCLoss                           | 0
604 | spec_augmentation                | SpectrogramAugmentation           | 0
605 | spec_augmentation.spec_cutout    | SpecCutout                        | 0
606 | _wer                             | WER                               | 0
-----------------------------------------------------------------------------------------
18.9 M    Trainable params
0         Non-trainable params
18.9 M    Total params

As the model starts training, you should see a progress bar per epoch.

Epoch 0: 100%|█████████████████████████████████████████████████████████████████████████████████████████| 35/35 [00:14<00:00,  2.40it/s, loss=62.5Epoch 0, global step 29: val_loss reached 307.90469 (best 307.90469), saving model to "/tlt-nemo/checkpoints/trained-model---val_loss=307.90-epoch=0.ckpt" as top 3
Epoch 1: 100%|█████████████████████████████████████████████████████████████████████████████████████████| 35/35 [00:14<00:00,  2.48it/s, loss=57.6Epoch 1, global step 59: val_loss reached 70.93443 (best 70.93443), saving model to "/tlt-nemo/checkpoints/trained-model---val_loss=70.93-epoch=1.ckpt" as top 3
Epoch 2: 100%|█████████████████████████████████████████████████████████████████████████████████████████| 35/35 [00:14<00:00,  2.42it/s, loss=55.5Epoch 2, global step 89: val_loss reached 465.39551 (best 70.93443), saving model to "/tlt-nemo/checkpoints/trained-model---val_loss=465.40-epoch=2.ckpt" as top 3
Epoch 3:  60%|█████████████████████████████████████████████████████▍                                   | 21/35 [00:09<00:06,  2.19it/s, loss=54.5]
...

At the end of training, TAO Toolkit will save the last checkpoint at the path specified by the experiment spec file before finishing.

[NeMo I 2021-01-20 22:38:48 train:120] Experiment logs saved to '.'
[NeMo I 2021-01-20 22:38:48 train:123] Trained model saved to './checkpoints/trained-model.tlt'
INFO: Internal process exited

Troubleshooting

  • Currently, only .wav audio files are supported.

  • If you are training on a non-English dataset and are consistently getting blank predictions, check that you have turned normalize_transcripts=False. By default, the data layers have normalization on and will get rid of non-English characters.

  • Similarly, if you are training on an English dataset with capital letters or additional punctuation, make sure that the data layer normalizes transcripts to lowercase, or that your custom vocabulary includes the additional valid characters.

  • If you consistently run into out-of-memory errors while training, consider adding a maximum length to your audio samples using max_duration.

Evaluating the Model

To run evaluation on a trained model checkpoint, use this command:

tao speech_to_text_citrinet evaluate -e <experiment_spec> \
                             -m <model_checkpoint> \
                             -g <num_gpus>

Required Arguments

  • -e: The experiment specification file to set up evaluation. This only needs a dataset config, as described in the “Dataset Configs” section.

  • -m: Path to the model checkpoint, which should be a .tlt file.

Optional Arguments

  • -g: The number of GPUs to be used in evaluation in a multi-GPU scenario (default: 1).

  • -r: The path to the results and log directory. Log files, checkpoints, etc., will be stored here.

  • -k: The key to encrypt the model.

  • Other arguments to override fields in the specification file.

Evaluation Procedure

At the start of evaluation, TAO Toolkit will print out a log of the experiment specification, including any parameters added or overridden via the command line. It will also show additional information, such as which GPUs are available, where logs will be saved, and how many hours are in the test dataset.

GPU available: True, used: True
TPU available: None, using: 0 TPU cores
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0,1,2]
[NeMo W 2021-01-20 22:58:19 exp_manager:375] Exp_manager is logging to ./, but it already exists.
[NeMo I 2021-01-20 22:58:19 exp_manager:186] Experiments will be logged at .
...
[NeMo I 2021-01-20 22:58:20 features:235] PADDING: 16
[NeMo I 2021-01-20 22:58:20 features:251] STFT using torch
[NeMo I 2021-01-20 22:58:22 collections:173] Dataset loaded with 130 files totaling 0.10 hours
[NeMo I 2021-01-20 22:58:22 collections:174] 0 files were filtered totaling 0.00 hours

Once evaluation begins, a progress bar will be shown to indicate how many batches have been processed. After evaluation, the test results will be shown.

Testing: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████| 5/5 [00:02<00:00,  2.43it/s]
--------------------------------------------------------------------------------
DATALOADER:0 TEST RESULTS
{'test_loss': tensor(68.1998, device='cuda:0'),
 'test_wer': tensor(0.9987, device='cuda:0')}
--------------------------------------------------------------------------------
[NeMo I 2021-01-20 22:58:24 evaluate:94] Experiment logs saved to '.'
INFO: Internal process exited

Troubleshooting

  • Currently, only .wav audio files are supported.

  • Filtering should be turned off for evaluation. Make sure that max_duration and min_duration are not set.

  • For best results, evaluation should be done on audio files with the same sample rate as the training data.

  • The model will only predict characters that were included in the training vocabulary. Make sure that the training and test vocabularies match, including normalization.

Fine-Tuning the Model

To fine-tune a model from a checkpoint, use the following command:

tao speech_to_text_citrinet finetune -e <experiment_spec> \
                             -m <model_checkpoint> \
                             -g <num_gpus>

Required Arguments

  • -e: The experiment specification file to set up fine-tuning. This requires the trainer, save_to, and optim configurations described in the “Training Process Configs” section, as well as finetuning_ds and validation_ds configs, as described in the “Dataset Configs” section. Additionally, if your fine-tuning dataset has a different vocabulary (i.e., set of labels) than the trained model checkpoint, you must also set change_vocabulary: true at the top level of your specification file.

  • -m: Path to the model checkpoint from which to fine-tune. Should be a .tlt file.

Optional Arguments

  • -g: The number of GPUs to be used for fine-tuning in a multi-GPU scenario (default: 1).

  • -r: The path to the results and log directory. Log files, checkpoints, etc., will be stored here.

  • -k: The key to encrypt the model.

  • Other arguments to override fields in the specification file.

Fine-Tuning Procedure

At the start of fine-tuning, TAO Toolkit will print out a log of the experiment specification, including any parameters added or overridden via the command line. It will also show additional information, such as which GPUs are available, where logs will be saved, and how many hours are in the fine-tuning and evaluation dataset.

When restoring from the checkpoint, it will then log the original datasets that the checkpoint model was trained and evaluated on.

If the vocabulary has been changed, the logs will show what the new vocabulary is.

[NeMo I 2021-01-20 23:33:12 finetune:110] Model restored from './checkpoints/trained-model.tlt'
[NeMo I 2021-01-20 23:33:12 ctc_models:247] Changed decoder to output to [' ', 'а', 'б', 'в', 'г', 'д', 'е', 'ё', 'ж', 'з', 'и', 'й', 'к', 'л', 'м', 'н', 'о', 'п', 'р', 'с', 'т', 'у', 'ф', 'х', 'ц', 'ч', 'ш', 'щ', 'ъ', 'ы', 'ь', 'э', 'ю', 'я'] vocabulary.
[NeMo I 2021-01-20 23:33:12 collections:173] Dataset loaded with 7242 files totaling 11.74 hours
[NeMo I 2021-01-20 23:33:12 collections:174] 0 files were filtered totaling 0.00 hours
[NeMo I 2021-01-20 23:33:12 collections:173] Dataset loaded with 7307 files totaling 12.56 hours
[NeMo I 2021-01-20 23:33:12 collections:174] 0 files were filtered totaling 0.00 hours

As with training, TAO Toolkit will log a full listing of the modules and submodules in the model, as well as the total number of trainable and non-trainable parameters in the model. See the Training section for more details on the parameter breakdowns.

    | Name                             | Type                              | Params
-----------------------------------------------------------------------------------------
0   | preprocessor                     | AudioToMelSpectrogramPreprocessor | 0
1   | preprocessor.featurizer          | FilterbankFeatures                | 0
2   | encoder                          | ConvASREncoder                    | 18.9 M
...
603 | decoder                          | ConvASRDecoder                    | 35.9 K
604 | decoder.decoder_layers           | Sequential                        | 35.9 K
605 | decoder.decoder_layers.0         | Conv1d                            | 35.9 K
606 | loss                             | CTCLoss                           | 0
-----------------------------------------------------------------------------------------
18.9 M    Trainable params
0         Non-trainable params
18.9 M    Total params

Note that if the vocabulary has been changed, the decoder may have a different number of parameters.

Fine-tuning on the new dataset should proceed afterwards as with normal training, with a progress bar per epoch and checkpoints saved to the specified directory.

Troubleshooting

  • Currently, only .wav audio files are supported.

  • We recommend using a lower learning rate for fine-tuning from a trained model checkpoint. A good criteria to start with is 1/100 of the original learning rate.

  • If the fine-tuning vocabulary is different from the original training vocabulary, you will need to set change_vocabulary=True.

  • You may see a dimensionality mismatch error (example below) or other hyperparameter mismatch error if your training checkpoint directory (i.e., the model you are loading with restore_from) and fine-tuning checkpoint directory are the same. Make sure they are distinct by using the -r flag (e.g., -r <new/log/dir>).

RuntimeError: Error(s) in loading state_dict for EncDecCTCModel:
`size mismatch for decoder.decoder_layers.0.weight: copying a param with shape torch.Size([29, 1024, 1]) from checkpoint, the shape in current model is torch.Size([35, 1024, 1]).`

Using Inference on a Model

To perform inference on individual audio files, use the following command:

tao speech_to_text_citrinet infer -e <experiment_spec> \
                          -m <model_checkpoint> \
                          -g <num_gpus>

Required Arguments

  • -e: The experiment specification file to set up inference. This spec file only needs a file_paths parameter that contains a list of individual file paths.

  • -m: Path to the model checkpoint. Should be a .tlt file.

Optional Arguments

  • -g: The number of GPUs to be used for inference in a multi-GPU scenario (default: 1).

  • -r: The path to the results and log directory. Log files, checkpoints, etc., will be stored here.

  • -k: The key to encrypt the model.

  • Other arguments to override fields in the specification file.

Inference Procedure

At the start of inference, TAO Toolkit will print out the experiment specification, including the audio filepaths on which inference will be performed.

When restoring from the checkpoint, it will then log the original datasets that the checkpoint model was trained and evaluated on. This will show the vocabulary that the model was trained on.

Train config :
manifest_filepath: /data/an4/train_manifest.json
batch_size: 32
sample_rate: 16000
labels:
- ' '
- a
- b
- c
...

Prediction results will be shown at the end of the log. Each prediction is preceded by the associated filename on the previous line.

[NeMo I 2021-01-21 00:22:00 infer:67] The prediction results:
[NeMo I 2021-01-21 00:22:00 infer:69] File: /data/an4/wav/an4test_clstk/fcaw/an406-fcaw-b.wav
[NeMo I 2021-01-21 00:22:00 infer:70] Predicted transcript: rubout g m e f three nine
[NeMo I 2021-01-21 00:22:00 infer:69] File: /data/an4/wav/an4test_clstk/fcaw/an407-fcaw-b.wav
[NeMo I 2021-01-21 00:22:00 infer:70] Predicted transcript: erase c q q f seven
[NeMo I 2021-01-21 00:22:00 infer:73] Experiment logs saved to '.'
INFO: Internal process exited

Troubleshooting

  • Currently, only .wav audio files are supported.

  • For best results, inference should be done on audio files with the same sample rate as the training data.

  • The model will only predict characters that were included in the training vocabulary. Make sure that the training and test vocabularies match, including normalization.

Model Export

You can export a trained ASR model to the Riva format, which contains all the model artifacts necessary for deployment to Riva Services. For more details about Riva, see this page.

To export an ASR model to the Riva format, use the following command:

tao speech_to_text_citrinet export -e <experiment_spec> \
                           -m <model_checkpoint> \
                           -r <results_dir> \
                           -k <encryption_key> \
                           export_format=RIVA

Required Arguments

  • -e: The experiment specification file for export. See the Export Spec File section below.

  • -m: Path to the model checkpoint to be exported. Should be a .tlt file.

Optional Arguments

  • -k: The encryption key.

  • -r: Path to the directory where results will be stored.

Export Spec File

The following is an example spec file for model export.

# Name of the .tlt EFF archive to be loaded/model to be exported.
restore_from: trained-model.tlt

# Set export format: RIVA
export_format: RIVA

# Output EFF archive containing model checkpoint and artifacts required for Riva Services
export_to: exported-model.riva

Parameter

Datatype

Description

Default

restore_from

string

Path to the pre-trained model to be exported.

trained_model.tlt

export_format

string

Export format.

N/A

export_to

string

Target path for the export model.

exported-model.riva