# Behavior traces — openai/whisper

> Canonical page: https://codevetter.com/unpack/whisper/behavior-traces

# Behavior traces — openai/whisper

Scanned commit 86098128c0b4f24f0e2aa2994de830614b474227 (2026-08-31). Generated by codevetter codevetter 1.12.0 on 2026-09-22.

Three end-to-end flows were traced. (1) CLI startup: cli() → argparse → load_model → per-file transcribe → writer. (2) Model loading: name → _MODELS URL or local path → _download with SHA256 → torch.load → Whisper(dims) → set_alignment_heads. (3) Transcription: mel → language detect → per-window decode_with_fallback → DecodingTask.run (encoder → language → _main_loop → finalize → rank) → timestamp segmentation → optional word-level DTW alignment → result dict → file writers.
- CLI flow: whisper/__main__.py → transcribe.cli() → argparse (~30 args, valid_model_name validates against available_models() or file existence) → os.makedirs(output_dir) → .en-model language guard → temperature ladder built from --temperature_increment_on_fallback → load_model(model_name, device, download_root=model_dir) → get_writer(output_format, output_dir) → for each audio: transcribe(model, path, temperature=..., **args) then writer(result, audio_path, **writer_args). — [whisper/__main__.py](https://github.com/openai/whisper/blob/86098128c0b4f24f0e2aa2994de830614b474227/whisper/__main__.py) [whisper/transcribe.py#L517-619](https://github.com/openai/whisper/blob/86098128c0b4f24f0e2aa2994de830614b474227/whisper/transcribe.py#L517-L619) [whisper/__init__.py#L98-100](https://github.com/openai/whisper/blob/86098128c0b4f24f0e2aa2994de830614b474227/whisper/__init__.py#L98-L100)
- CLI resilience: each file's transcribe+write is wrapped in try/except that prints the traceback and `Skipping <path> due to ...`, so one bad file doesn't abort the batch. — [whisper/transcribe.py#L613-619](https://github.com/openai/whisper/blob/86098128c0b4f24f0e2aa2994de830614b474227/whisper/transcribe.py#L613-L619)
- Model-load flow: load_model picks device (cuda if available) and download_root (~/.cache/whisper or $XDG_CACHE_HOME/whisper) → _download(url, root, in_memory) creates root, checks existing file's SHA256 (re-downloads on mismatch with a warning), streams via urllib with tqdm progress, verifies hash again post-download → torch.load(weights_only=True on torch>=1.13) → ModelDimensions(**checkpoint['dims']) → Whisper(dims) → load_state_dict → set_alignment_heads for official models. — [whisper/__init__.py#L103-161](https://github.com/openai/whisper/blob/86098128c0b4f24f0e2aa2994de830614b474227/whisper/__init__.py#L103-L161)
- Transcribe flow (in order): dtype/fp16 resolution with CPU-fp16 downgrade warning → log_mel_spectrogram(audio, n_mels, padding=N_SAMPLES) → language detection via model.detect_language on first 30 s when unset → get_tokenizer → clip_timestamps → seek_clips pairs → while clip_idx < len(seek_clips): slice mel_segment, pad_or_trim, build prompt, decode_with_fallback → split tokens on consecutive timestamps into new_segment dicts → optional add_word_timestamps → append to all_segments/all_tokens → return dict. — [whisper/transcribe.py#L127-177](https://github.com/openai/whisper/blob/86098128c0b4f24f0e2aa2994de830614b474227/whisper/transcribe.py#L127-L177) [whisper/transcribe.py#L263-514](https://github.com/openai/whisper/blob/86098128c0b4f24f0e2aa2994de830614b474227/whisper/transcribe.py#L263-L514)
- decode_with_fallback retries the window at each temperature in the ladder, dropping beam_size/patience when t>0 and best_of when t==0, and marks failure when compression_ratio > 2.4 or avg_logprob < -1.0 — unless no_speech_prob > 0.6 together with low logprob, which instead classifies the segment as silence. — [whisper/transcribe.py#L184-224](https://github.com/openai/whisper/blob/86098128c0b4f24f0e2aa2994de830614b474227/whisper/transcribe.py#L184-L224) [whisper/transcribe.py#L298-310](https://github.com/openai/whisper/blob/86098128c0b4f24f0e2aa2994de830614b474227/whisper/transcribe.py#L298-L310)
- DecodingTask.run order: decoder.reset() → _get_audio_features (encoder forward, fp16 cast) → tokens = initial_tokens repeated per audio → _detect_language (optionally writes language token into sot_index+1, short-circuits for task='lang_id') → repeat_interleave by n_group (beam_size or best_of) → _main_loop → reshape (n_audio, n_group) → decoder.finalize → slice tokens between sample_begin and first EOT → sequence_ranker.rank → build DecodingResult list with avg_logprob, no_speech_prob, compression_ratio. — [whisper/decoding.py#L712-789](https://github.com/openai/whisper/blob/86098128c0b4f24f0e2aa2994de830614b474227/whisper/decoding.py#L712-L789)
- Word-timestamp flow: add_word_timestamps → find_alignment (hooks on cross_attn capture QK, model() under disable_sdpa, weights over alignment_heads → median_filter → mean → dtw(-matrix) → split_to_word_tokens → WordTiming list) → sentence-boundary duration clamp → merge_punctuations → distribute words into segment['words'] with seek-derived time_offset and segment start/end reconciliation heuristics. — [whisper/timing.py#L163-242](https://github.com/openai/whisper/blob/86098128c0b4f24f0e2aa2994de830614b474227/whisper/timing.py#L163-L242) [whisper/timing.py#L279-388](https://github.com/openai/whisper/blob/86098128c0b4f24f0e2aa2994de830614b474227/whisper/timing.py#L279-L388)
- Persistence flow: writer(result, audio_path) → ResultWriter.__call__ → output_path = output_dir/<basename>.<ext> → open utf-8 → write_result; format 'all' returns a closure invoking all six writers; subtitle writers produce `start --> end\ntext` blocks via iterate_result. — [whisper/utils.py#L85-107](https://github.com/openai/whisper/blob/86098128c0b4f24f0e2aa2994de830614b474227/whisper/utils.py#L85-L107) [whisper/utils.py#L238-263](https://github.com/openai/whisper/blob/86098128c0b4f24f0e2aa2994de830614b474227/whisper/utils.py#L238-L263) [whisper/utils.py#L306-329](https://github.com/openai/whisper/blob/86098128c0b4f24f0e2aa2994de830614b474227/whisper/utils.py#L306-L329)
- Word-anomaly / hallucination-skip flow inside the seek loop: is_segment_anomaly scores words (prob<0.15, duration<0.133 s or >2 s) on the first 8 non-punctuation words; when hallucination_silence_threshold is set and an anomalous segment is surrounded by silence, the seek jumps past it and trailing current_segments are dropped. — [whisper/transcribe.py#L315-337](https://github.com/openai/whisper/blob/86098128c0b4f24f0e2aa2994de830614b474227/whisper/transcribe.py#L315-L337) [whisper/transcribe.py#L418-471](https://github.com/openai/whisper/blob/86098128c0b4f24f0e2aa2994de830614b474227/whisper/transcribe.py#L418-L471)

Every claim links to its cited files at the pinned commit. Citations identify source evidence; they do not establish runtime correctness. Independent CodeVetter scan — not reviewed by upstream maintainers.

## Public product links

- [CodeVetter](https://codevetter.com/)
- [Download](https://codevetter.com/download)
- [Documentation](https://codevetter.com/docs/)
- [Source](https://github.com/Codevetter/codevetter)