OLMoASROptions

Options & Configuration · AiDotNet.SpeechRecognition.LLMIntegrated

Options for OLMoASR: Whisper’s encoder-decoder trained on the open OLMoASR-Mix.

Inheritance: ModelOptions → OLMoASROptions

For Beginners

Defaults are the released OLMoASR-large.en-v2 (1.5B parameters): Whisper large’s geometry (80-band log-mels over 30-second windows, width 1,280, 20 heads, 32 encoder and 32 decoder layers) with the English-only models’ 51,864-token GPT-2 vocabulary. The smaller releases are tiny.en (384, 6 heads, 4 + 4 layers), base.en (512, 8, 6 + 6), small.en (768, 12, 12 + 12) and medium.en (1,024, 16, 24 + 24).

Constructors

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

Properties

Property Summary
OnnxModelOptions OnnxOptions { get; set; } Gets or sets the ONNX Runtime settings used when the model is loaded from an ONNX file.
int MaxTextLength { get; set; } The longest transcript generated, in tokens (224, half the decoder’s 448-token context, as Whisper’s sample_len).
int ModelDim { get; set; } The encoder’s and decoder’s width (1,280 for large).
int NumAttentionHeads { get; set; } Attention heads in the encoder and the decoder (20 for large).
int NumDecoderLayers { get; set; } Decoder layers (32 for large).
int NumEncoderLayers { get; set; } Encoder layers (32 for large).
int NumMels { get; set; } Log-mel bands (80, as every English-only Whisper geometry).
int SampleRate { get; set; } Gets or sets the sample rate, in Hz, that input audio is expected at.
int VocabSize { get; set; } The vocabulary (51,864, the English-only models’ GPT-2 encoding with Whisper’s special tokens).
string Language { get; set; } Gets or sets the language code used for speech processing.
string ModelPath { get; set; } Gets or sets the path to the model weights.
string[] Vocabulary { get; set; } Optional vocabulary mapping token ids to text when the vocabulary is not Whisper’s English one.