Whisper API

Converts audio files (MP3, WAV, M4A) and video files (MP4) to text using OpenAI's Whisper model deployed as a hosted REST API. The service automatically detects the spoken language from audio content and transcribes across 98+ languages without requiring explicit language specification. Transcription requests are processed asynchronously with real-time progress tracking via dashboard, and files are automatically deleted after 24 hours while transcripts persist indefinitely in user accounts.

Exposes multiple Whisper model size variants (including 'large-v2' and smaller options) as selectable parameters in API requests, allowing users to explicitly choose between accuracy and inference speed. Larger models provide higher accuracy but consume more credits and take longer to process; smaller models process faster with lower credit cost but reduced accuracy. The service claims to transform 10 minutes of audio to text in under a minute using optimized inference, though specific latency benchmarks per model size are not published.

Optionally identifies and separates speech from multiple speakers in a single audio file, labeling transcript segments with speaker identities (e.g., 'Speaker 1', 'Speaker 2'). Speaker diarization is implemented as an optional feature that increases the credit cost of transcription; the exact credit multiplier or cost formula is not documented. This capability enables meeting transcripts, interview recordings, and multi-speaker content to be transcribed with speaker attribution without manual post-processing.

Generates transcriptions in six distinct output formats (plain text, JSON with timestamps, SRT subtitles, VTT subtitles, DOCX, PDF) from a single audio/video input without requiring separate processing or format conversion steps. The API accepts a 'format' parameter specifying desired output, and the service handles format conversion server-side. Timestamp information is embedded in structured formats (JSON, SRT, VTT) enabling subtitle synchronization with video playback.

Implements a credit-based pricing model where each transcription consumes a variable number of credits determined by model size, speaker diarization, and file size. Users receive a cost preview showing exact credit consumption before confirming transcription, enabling informed decisions about feature selection and model size. Credits are purchased in tiered bundles ($5 for 20 credits up to $0.10/credit at 1000+ volume) and never expire, eliminating time-based pressure to consume credits. Free tier provides 5 daily transcription credits without requiring payment.

Processes transcription requests asynchronously via REST API, returning immediately with a job ID while transcription occurs server-side. Users can monitor transcription progress in real-time via a web dashboard showing processing status, estimated completion time, and final results. This non-blocking approach enables applications to submit multiple transcription requests without waiting for individual completions, and the dashboard provides visibility into queue status and processing metrics.

Automatically deletes uploaded audio/video files from the service after 24 hours while preserving transcription text indefinitely in user accounts. This design balances privacy (source files not permanently stored) with usability (transcripts remain accessible for reference, editing, and export). Users must download transcripts or export results within 24 hours if they need to preserve the original file, but can access transcription text from their account indefinitely.

Accepts remote URLs pointing to audio/video files instead of requiring local file uploads, enabling transcription of content hosted on external servers (e.g., CDNs, cloud storage, streaming platforms). The service downloads the file from the URL, processes transcription, and applies the same 24-hour deletion policy. This capability eliminates the need to download large files locally before uploading, reducing bandwidth and enabling direct transcription of hosted content.