whisper0.1.0 package

Apply BPE Merges

Get Audio Duration

Convert Audio to Mel Spectrogram

Convert Byte to BPE Token

Clean Transcribed Text

Compute STFT Magnitude

Copy Weight if Exists

Create Decoder from Config

Create Encoder from Config

Create Mel Filterbank (Fallback)

Decode BPE Bytes Back to Text

Decode Timestamp Token

Download Tokenizer Files from HuggingFace

Download Model from HuggingFace

Ensure Tokenizer Files are Downloaded

Extract Segments with Timestamps

Get Initial Decoder Tokens

Get Model Cache Path

Get Path to Model Weights

Greedy Decoding

Convert Hz to Mel Scale

Check if Token is Timestamp

List Downloaded Models

List Available Models

Load Added Tokens from HuggingFace

Load and Preprocess Audio

Load Decoder Weights

Load Encoder Weights

Load Pre-computed Mel Filterbank

Load Whisper Model

Load Weights from Safetensors

Convert Mel Scale to Hz

Check if Model is Downloaded

Pad or Trim Audio to Fixed Length

Parse Device Argument

Parse Dtype Argument

Split Long Audio into Chunks

Decode Token IDs to Text

Encode Text to Token IDs

Transcribe Single Chunk

Transcribe Long Audio

Whisper Transcription

Whisper Encoder

Whisper Model Configurations

Whisper Decoder

Text Decoder

Device and Dtype Management

Get Default Dtype

Encoder Layer

Audio Encoder

Get Language Token ID

Whisper Model

Audio Preprocessing for Whisper

Special Token IDs

Whisper BPE Tokenizer