deepspeech 命令

常用示例

说明

DeepSpeech is an open-source speech-to-text engine based on deep learning. It uses an end-to-end neural network architecture to convert audio into text transcriptions. The system requires a trained model and optionally an external scorer (language model) for improved accuracy. Pre-trained English models are available, and the toolkit supports training custom models for other languages or domains. Audio input must be 16kHz, 16-bit, mono WAV format. The tool supports both batch transcription of files and real-time streaming transcription through its API.

参数

--model _file_

Path to the model file (.pbmm or .tflite).

--scorer _file_

Path to external scorer/language model.

--audio _file_

Audio file to transcribe (16kHz, 16-bit, mono WAV).

--extended

Output word timing and confidence.

--json

Output results as JSON.

--beam_width _n_

Beam width for the CTC decoder.

--lm_alpha _value_

Language model weight. If not specified, uses default from the scorer package.

--lm_beta _value_

Word insertion bonus. If not specified, uses default from the scorer package.

--candidate_transcripts _n_

Number of candidate transcripts to include in JSON output (default: 3).

--hot_words _words_

Hot-words and their probability boosts.

--version

Print version and exit.