Term
Whisper initial_prompt
The `initial_prompt` parameter lets you pass Whisper a keyword list or sample sentence before transcription. The model treats this text as context and recognises the included terms far more reliably.
Whisper initial_prompt — explained in more detail
initial_prompt is an optional text parameter when calling a Whisper model (locally via whisper/faster-whisper, or through the OpenAI API). Whisper interprets the content as if it had just heard that text immediately before the audio it is about to transcribe. Effect: proper names, technical terms and specific spellings contained in the hint are preferred during decoding because they sit inside the current language and vocabulary context. It is not fine-tuning and not real training — the model weights remain unchanged. The hint only affects the single call and requires no additional infrastructure.
Example / Practical context
Anyone speaking regularly about in-house product names, tooling or code feeds Whisper a short keyword list — typically ten to fifteen terms, comma- or sentence-separated. “Debbie Runner” reliably becomes “DEVI Runner”; “Heiko” turns back into “Haiku”. With Whisper-small the list is hard-capped: beyond ~15 terms general transcription quality degrades because the model loses focus. Whisper-medium and -large tolerate more (around 20–25 terms), but the parameter is not arbitrarily scalable.
Distinction from related concepts
Fine-tuning permanently alters the model weights and requires training data, GPU time and evaluation — initial_prompt is merely a temporary, per-request context hint. A downstream correction layer (e.g. an LLM cleaning up Whisper output) handles word splits and hallucinations that a glossary hint cannot fix. The language parameter and condition_on_previous_text are other Whisper knobs, but not substitutes — they control language and context window, not domain vocabulary.
Entdecke mehr
Headless without an API bill — how do you reach the best AI models for automation in 2026?
Provider comparison mid-2026: who has a headless mode, whose subscription still covers it — and why BYOK is the most stable foundation.
GlossarEU AI Act (KI-Verordnung)
Regulation (EU) 2024/1689 is the EU's first comprehensive law on artificial intelligence. It regulates AI systems by risk in four tiers, from prohibited practices to minimal risk, and sets obligations for providers and deployers.
LexikonFunction Calling / Tool Use
How an LLM uses tools: define a tool as a schema, the model picks the function and arguments, the result returns to the chat — the basis of every agent.