Heads up This site is currently under heavy development.
← all tools
◆ AI Model & Data Infrastructure

Whisper

open-source
source ↗ rss ↗

Summary

Whisper is an open-source speech recognition model, distributed as a Python package, that transcribes and translates audio and identifies spoken language across multiple languages. It runs from the command line via `pip install openai-whisper` (with `ffmpeg` as a system dependency), or can be called as a library in a Python pipeline, which fits developers and ML engineers building transcription or translation into an application rather than someone wanting a hosted service. Its multitask Transformer design replaces what was previously a chain of separate speech-processing tools, handling recognition, translation, and language identification in one model rather than one pipeline stage each. It sits in the ai-infrastructure space around speech-to-text, with options like word-level timestamps and hallucination filtering for subtitle-style output. The project has ten tracked releases, though the most recent one landed 421 days ago.

What Whisper answers

Which languages does it actually handle well, or does it favor English?

trained on a broad multilingual dataset, but recognition and translation quality varies by language since English speech dominated the training data

Does it run on my hardware, or do I need a GPU?

runs on CPU or GPU through PyTorch, but larger model sizes are slow enough on CPU that most real use expects a GPU

How do I keep output in sync with long recordings for subtitles?

word-level timing and controls that skip likely-hallucinated silent stretches keep breaks aligned to actual speech

Is this still being actively developed if something breaks?

ten releases are on record but the latest is over a year old, so fixes may land slowly or fall to the community

all 4 features, with the evidence for each →

Features

4 capabilities · 4 backed by code, an API document or a real run

Built from everything we hold on Whisper — every release we have summarised, its product documentation and how that documentation has changed, its README, its command-line surface and API, and runs we performed ourselves. Dates are when we first saw a capability, not when the vendor introduced it.

Capability area
All capabilities 4 capabilities
Word-level timestamp segmentation verified Controls how transcription output is broken into lines and segments when word-level timestamps are enabled, via configurable limits on line width, line count, and words per line. 3 other sources · first seen Sep 2023

command line

  • --max_words_per_line — (requires --word_timestamps True, no effect with --max_line_width) the maximum number of words in a segment v20231105 · Nov 2023 · command-line history
  • --max_line_width — (requires --word_timestamps True) the maximum number of characters in a line before breaking the line v20230918 · Sep 2023 · command-line history
  • --max_line_count — (requires --word_timestamps True) the maximum number of lines in a segment v20230918 · Sep 2023 · command-line history
Persistent initial prompt verified Optionally prepends the initial prompt to every internal decode call throughout transcription, trading off against the effectiveness of conditioning on previous text. 1 other source · first seen Jun 2025

command line

  • --carry_initial_prompt — if True, prepend initial_prompt to every internal decode() call. May reduce the effectiveness of condition_on_previous_text v20250625 · Jun 2025 · command-line history
Clip-range transcription verified Restricts transcription to specified time ranges within a file by accepting a comma-separated list of start/end timestamps in seconds. 1 other source · first seen Sep 2024

command line

  • --clip_timestamps — comma-separated list start,end,start,end,... timestamps (in seconds) of clips to process, where the last end timestamp defaults to the end of the file v20240927 · Sep 2024 · command-line history
Hallucination silence suppression verified Skips silent periods that exceed a configurable duration threshold when a possible hallucination is detected, requiring word-level timestamps to be enabled. 1 other source · first seen Sep 2024

command line

  • --hallucination_silence_threshold — (requires --word_timestamps True) skip silent periods longer than this threshold (in seconds) when a possible hallucination is detected v20240927 · Sep 2024 · command-line history
Capability
Evidence

Lines in monospace are the tool's own words — help text parsed from its source, or an endpoint from its API document. Everything else is our summary of a dated release or documentation change, linked back to the source it came from.

Release history

Nothing shipped since we started watching. Releases and docs updates land here as they happen.

watching since Aug 3, 2026

my-toolchain — 0 tools
paste an install list to detect your tools

A brew list, a Brewfile, requirements.txt, a Dockerfile — or just the product names, free-form. Nothing leaves your browser.

    browse all tools →