Implementation:Openai Whisper Get Writer

Overview

get_writer() is a factory function that returns a callable writer for serializing Whisper transcription results into standard output formats. It is the public API for output formatting, backed by a hierarchy of ResultWriter subclasses.

Source

• File: whisper/utils.py:L296-318 (get_writer), whisper/utils.py:L85-106 (ResultWriter base), whisper/utils.py:L109-116 (WriteTXT), whisper/utils.py:L238-248 (WriteVTT), whisper/utils.py:L251-262 (WriteSRT), whisper/utils.py:L265-284 (WriteTSV), whisper/utils.py:L287-293 (WriteJSON)
• Import: from whisper.utils import get_writer
• Repository: https://github.com/openai/whisper

Signature

def get_writer(output_format: str, output_dir: str) -> Callable[[dict, TextIO, dict], None]:

Parameters

Parameter	Type	Description
output_format	str	One of "txt", "vtt", "srt", "tsv", "json", or "all"
output_dir	str	Directory path where output files will be written

Inputs and Outputs

• Inputs: Transcription result dictionary with keys: text, segments, language
• Outputs: Written file(s) in the specified format in the output directory

Writer Hierarchy

All writers inherit from ResultWriter, which provides the base __call__ method for file handling. Each subclass implements the specific formatting logic:

Writer Class	Format	Extension	Description
WriteTXT	Plain text	.txt	Text content only, no timestamps or metadata
WriteVTT	WebVTT	.vtt	Web subtitle format with WEBVTT header and HH:MM:SS.mmm timestamps
WriteSRT	SubRip	.srt	Standard subtitle format with sequential indices and HH:MM:SS,mmm timestamps
WriteTSV	Tab-separated	.tsv	Tabular format with millisecond timestamps
WriteJSON	JSON	.json	Full structured output including all metadata

When output_format="all", the function returns a composite writer that invokes all five format writers, producing one file of each type.

Usage Examples

from whisper.utils import get_writer

# Get a single writer
writer = get_writer("srt", "./output")
writer(result, audio_path)  # Writes output/audio.srt

# Get all writers
writer = get_writer("all", "./output")
writer(result, audio_path)  # Writes .txt, .vtt, .srt, .tsv, .json

Output Format Details

TXT Output

Writes the text from each segment, concatenated. No timing information.

VTT Output

WEBVTT

00:00:00.000 --> 00:00:02.400
Hello, this is a test.

00:00:02.400 --> 00:00:05.800
This is the second segment.

SRT Output

1
00:00:00,000 --> 00:00:02,400
Hello, this is a test.

2
00:00:02,400 --> 00:00:05,800
This is the second segment.

TSV Output

start	end	text
0	2400	Hello, this is a test.
2400	5800	This is the second segment.

JSON Output

Serializes the full result dictionary to JSON, including all segment metadata (tokens, probabilities, compression ratios).

Key Notes

• The output_dir must exist before calling the writer.
• Output filenames are derived from the audio file path (e.g., speech.mp3 produces speech.srt).
• The ResultWriter base class handles file creation and encoding (UTF-8).
• The "all" option is useful for archival purposes when multiple consumers need different formats.

See Also

• Principle:Openai_Whisper_Output_Formatting