Audio Chunk Preparer — bounded WAV parts for transcription workflows

Prepare MP3, WAV, FLAC or Ogg audio for your own speech-to-text service. Get ordered 16 kHz mono PCM WAV files with exact decoded-sample ranges, overlap and byte counts. No transcription API key, model call or transcript is included.

Useful when your workflow can already transcribe audio but needs smaller files and source-time offsets. Free FFmpeg scripts are a good alternative if you already maintain a media-processing server. This Actor packages that step for Apify/API workflows.

Input

Supply one accessible Apify record URL in audioFiles, or raw audioBase64 for files up to 6,000,000 bytes. Do not supply both; clear the base64 example before entering a URL.

For small local files without an API token: use the free Console input generator. Run python prepare_audio.py recording.wav --console-json audio-input.json, then paste the generated JSON into this Actor's JSON input editor. This is a local conversion; it does not upload or run anything. Keep the generated file private because it contains your audio. Start the Actor separately when ready.

For larger files: upload with your own Apify client and obtain a record-specific signed URL using get_record_public_url. The Python workflow example demonstrates this route. A normal private Console upload URL can fail with UPLOAD_NOT_ACCESSIBLE; the native upload widget has therefore been removed. Do not repeatedly retry an unchanged inaccessible URL.

URLs must point directly to an Apify key-value-store record. Arbitrary external hosts, redirects, account-token query strings and remote playlists are rejected. Treat signed links as confidential. No account credentials are forwarded to downloads.

{"audioFiles":["https://api.apify.com/v2/key-value-stores/YOUR_STORE/records/audio.wav?signature=YOUR_RECORD_SIGNATURE"],"chunkSeconds":300,"overlapSeconds":1,"maxChunkBytes":20000000}

Limits: one input up to 64 MiB, decoded duration up to 30 minutes, 120 output parts maximum. chunkSeconds: integer 5–600; overlapSeconds: integer 0–5 and smaller than the effective chunk; maxChunkBytes: integer 160044–24000000. The byte cap can shorten your selected chunk duration. Large input belongs in an uploaded binary record, not base64 inside JSON: Apify Actor input has a separate 9 MB limit.

Supported containers: WAV, MP3, FLAC, Ogg. M4A, MP4/video and remote stream playlists are not supported in this version. Only the first audio stream is processed. Overlong recordings are rejected, not silently truncated.

Output and workflow

The dataset contains one manifest. Each chunks entry has index, recordKey, bytes, start_sample, end_sample_exclusive, start_seconds, end_seconds, duration_seconds, and overlap_with_previous_seconds.

1. Start the Actor and wait for success.
2. Read its dataset manifest and the run's defaultKeyValueStoreId.
3. Download each recordKey from that store using your own Apify client/authentication.
4. Send the WAV binaries to your chosen transcription provider in index order.
5. Add start_seconds to provider timestamps. Account for overlap before joining transcripts: repeated audio may create repeated words.

Output WAV files are stored in your run's key-value store. No public download token is inserted into the manifest. Treat input/output storage as your responsibility: normal Apify retention applies; delete sensitive recordings when no longer needed. Temporary decoder files are removed after decoding. No audio is sent to a transcription service by this Actor.

Price

USD 0.03 per successfully prepared input recording, not per chunk. Platform usage included in the advertised event price. No start or dataset event charge. Rejected input does not trigger the preparation event. Other services you choose and storage retained outside the run can have their own costs; check your account terms.

Limits and quality

This is fixed-duration/byte splitting, not silence detection or speaker diarization. Cuts can cross words. Overlap can help your downstream workflow but does not guarantee transcript quality. Mono mixing and 16 kHz resampling change audio and can lose information; keep the original when quality matters. Sample positions refer to the decoded audio timeline. Compressed-format priming or container timestamps may differ. No claim of model-specific compatibility or accuracy is made without testing your provider.

Errors include INVALID_OR_UNSUPPORTED_AUDIO, AUDIO_DURATION_LIMIT, INPUT_SIZE_LIMIT, TOO_MANY_CHUNKS, UPLOAD_NOT_ACCESSIBLE, USE_ONE_APIFY_UPLOAD_URL, and SPENDING_LIMIT. For too many parts, increase the duration/byte cap or reduce overlap. For inaccessible upload, use one directly readable Apify binary record URL with an optional Apify record signature (not an account token). Do not keep retrying unchanged invalid input.

Test scope

22 local synthetic codec, boundary, rejection and billing tests passed. Private Apify tests verified MP3 output, a signed 30-minute WAV upload with seven downloadable parts, one charge event per success and zero preparation events for invalid audio. These were owner tests, not evidence of customer payments. No real customer recording or paid transcription service was used. Report issues with synthetic audio, settings and error code only; never post private recordings, tokens or signed links.

Built by Agent Tools with AI assistance. FFmpeg is executed as a separate process via imageio-ffmpeg; upstream notices are included. Other Agent Tools · Workflow examples.