Class WhisperTokenizer

Namespace

AiDotNet.Audio.Whisper

Assembly

AiDotNet.dll

Tokenizer for Whisper speech recognition model.

public class WhisperTokenizer

Inheritance

object

WhisperTokenizer

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Remarks

Whisper uses a special tokenizer with BPE (Byte Pair Encoding) and special tokens for controlling transcription behavior (language, task, timestamps).

For Beginners: A tokenizer converts text to numbers (tokens) and back. Whisper's tokenizer has special tokens for:

• Language codes (to specify which language to transcribe)
• Task tokens (transcribe vs translate)
• Timestamp tokens (for word-level timing)

Properties

EndOfText

Gets the end of text token ID.

public int EndOfText { get; }

Property Value

int

NoSpeechToken

Gets the no speech token ID.

public int NoSpeechToken { get; }

Property Value

int

NoTimestampsToken

Gets the no timestamps token ID.

public int NoTimestampsToken { get; }

Property Value

int

StartOfTranscript

Gets the start of transcript token ID.

public int StartOfTranscript { get; }

Property Value

int

SupportedLanguages

Gets all supported language codes.

public static IReadOnlyList<string> SupportedLanguages { get; }

Property Value

IReadOnlyList<string>

TranscribeToken

Gets the transcribe task token ID.

public int TranscribeToken { get; }

Property Value

int

TranslateToken

Gets the translate task token ID.

public int TranslateToken { get; }

Property Value

int

Methods

Decode(IEnumerable<long>)

Decodes a sequence of token IDs to text.

public string Decode(IEnumerable<long> tokenIds)

Parameters

tokenIds IEnumerable<long>

The token IDs to decode.

Returns

string

The decoded text.

Remarks

This is a simplified decoder. A full implementation would use the actual BPE vocabulary from the Whisper model.

Encode(string)

Encodes text to token IDs.

public List<long> Encode(string text)

Parameters

text string

The text to encode.

Returns

List<long>

The encoded token IDs.

Remarks

This is a placeholder. A full implementation would use BPE encoding.

GetLanguageToken(string)

Gets the token ID for a language code.

public int GetLanguageToken(string languageCode)

Parameters

languageCode string

Two-letter language code (e.g., "en", "es").

Returns

int

The token ID for the language.

GetTimeFromToken(int)

Converts a timestamp token ID to time in seconds.

public double GetTimeFromToken(int tokenId)

Parameters

tokenId int

The timestamp token ID.

Returns

double

Time in seconds.

GetTimestampToken(double)

Gets the timestamp token ID for a given time in seconds.

public int GetTimestampToken(double timeSeconds)

Parameters

timeSeconds double

Time in seconds (must be a multiple of 0.02).

Returns

int

The timestamp token ID.

IsSpecialToken(int)

Checks if a token ID is a special token.

public bool IsSpecialToken(int tokenId)

Parameters

tokenId int

The token ID to check.

Returns

bool

True if the token is a special token.

IsTimestampToken(int)

Checks if a token ID is a timestamp token.

public bool IsTimestampToken(int tokenId)

Parameters

tokenId int

The token ID to check.

Returns

bool

True if the token is a timestamp token.