AzureSpeechSTTOptions
Options & Configuration · AiDotNet.SpeechRecognition.ProprietaryAPI
Options for the nearest disclosed architecture to Azure Speech, Microsoft’s streaming Conformer transducer (Chen et al. 2021).
Inheritance: ModelOptions → AzureSpeechSTTOptions
For Beginners
These options set the size of the encoder and decoder, the feature front end and the training augmentation.
How It Works
See AzureSpeechSTT for which defaults the vendor or paper discloses and which the stand-in infers.
Constructors
| Constructor | Summary |
|---|---|
AzureSpeechSTTOptions() |
Initializes a new instance with default values. |
AzureSpeechSTTOptions(AzureSpeechSTTOptions other) |
Initializes a new instance by copying from another instance. |
Properties
| Property | Summary |
|---|---|
OnnxModelOptions OnnxOptions { get; set; } |
ONNX runtime options. |
double RopeTheta { get; set; } |
Rotary base standing in for the paper’s relative position encoding. |
double TimeMaskRatio { get; set; } |
Largest time mask as a fraction of the utterance (pS = 0.05). |
int AttentionChunkSize { get; set; } |
Encoder frames (40 ms each) per attention chunk: 720 ms, the paper’s 24 frames of 30 ms (section 4.2.2). |
int AttentionLeftChunks { get; set; } |
Chunks of history each chunk also attends to: 2.16 s, the nearest whole number of chunks above the paper’s 60 frames (1.8 s). |
int ConvKernel { get; set; } |
Depthwise convolution kernel (section 4.1: 3). |
int EncoderDim { get; set; } |
Conformer width (section 4.1: 640). |
int FeedForwardUnits { get; set; } |
Hidden units of each feed-forward module (1,024, the Transformer transducer’s, section 4.1). |
int FrequencyMaskWidth { get; set; } |
Largest frequency mask, in mel bins (F = 27). |
int FrequencyMasks { get; set; } |
SpecAugment frequency masks (Park et al. |
int JointDim { get; set; } |
Width of the joint network (the encoder width; not stated). |
int MaxAudioLengthSeconds { get; set; } |
Longest audio a transcription covers, in seconds; longer input is truncated. |
int NumAttentionHeads { get; set; } |
Attention heads per block (not stated by the paper). |
int NumEncoderLayers { get; set; } |
Conformer blocks (18, the depth section 4.1 gives the Transformer transducer C-T is matched to). |
int NumMels { get; set; } |
Log-mel bins (80 over 25 ms windows with a 10 ms stride). |
int PredictionDim { get; set; } |
LSTM units of the prediction network (section 4.1: 720). |
int PredictionLayers { get; set; } |
LSTM layers of the prediction network (section 4.1: 2). |
int SampleRate { get; set; } |
Audio sample rate in Hz. |
int TimeMasks { get; set; } |
SpecAugment time masks (10). |
int VocabSize { get; set; } |
Output vocabulary size, blank (token 0) included: 4,000 sentence pieces (section 4.1) and the blank. |
string Language { get; set; } |
Language reported for transcriptions. |
string ModelPath { get; set; } |
Path to an exported ONNX model, when one is used. |
string[] Vocabulary { get; set; } |
The text of each token; tokens past the list decode as Unicode code points, and tokens in angle brackets decode to nothing. |