OmnilangualASROptions

Options & Configuration · AiDotNet.SpeechRecognition.Multilingual

Options for Omnilingual ASR’s CTC recognizer.

Inheritance: ModelOptions → OmnilangualASROptions

For Beginners

These options configure the Omnilingual ASR recognizer. The model listens to the raw audio waveform (not a spectrogram) and outputs one character distribution per 20 ms frame.

How It Works

Omnilingual ASR (Omnilingual ASR team, Meta 2025) pre-trains wav2vec 2.0 encoders of 0.3B to 7B parameters on 4.3M hours in over 1,600 languages; its CTC models put a single linear layer on the encoder and train the whole model with a CTC loss. The defaults are the 300M model (Table 4: 24 layers, 1024 wide, 4096 feed-forward, 16 heads) in fairseq2’s large_lv60k layout (layer-norm feature encoder with conv bias, stable pre-norm Transformer), with the released CTC configuration’s regularization: no dropout on hidden states or attention, 0.1 inside each feed-forward block, LayerDrop 0.1, and no masking.

Constructors

Constructor Summary
OmnilangualASROptions() Initializes a new instance with default values.
OmnilangualASROptions(OmnilangualASROptions other) Initializes a new instance by copying from another instance.

Properties

Property Summary
OnnxModelOptions OnnxOptions { get; set; } ONNX runtime options for the ONNX constructor.
bool FreezeFeatureEncoder { get; set; } Whether fine-tuning keeps the convolutional feature encoder frozen.
double ActivationDropout { get; set; } Dropout inside each feed-forward block.
double DropoutRate { get; set; } Dropout on the Transformer’s hidden states and attention weights.
double FeatureProjectionDropout { get; set; } Dropout after the feature projection.
double FinalDropout { get; set; } Dropout before the CTC head.
double LayerDrop { get; set; } Probability of skipping each Transformer layer during training.
double MaskTimeProbability { get; set; } SpecAugment time-mask probability during training.
int EncoderDim { get; set; } Transformer width (hidden size).
int FeedForwardDim { get; set; } Feed-forward width of each Transformer layer.
int MaxAudioLengthSeconds { get; set; } Longest audio, in seconds, a transcription reads; longer input is truncated.
int MaxTextLength { get; set; } Most tokens a transcription emits.
int NumAttentionHeads { get; set; } Attention heads per layer.
int NumEncoderLayers { get; set; } Number of Transformer layers.
int PositionalConvGroups { get; set; } Groups of the convolutional positional embedding.
int SampleRate { get; set; } Audio sample rate in Hz.
int VocabSize { get; set; } CTC vocabulary size, blank (token 0) included; tokens above 0 decode as Unicode code points.
string Language { get; set; } Language reported for a transcription when none is given.
string ModelPath { get; set; } Path of an exported ONNX model, for the ONNX constructor.