Workspace
Search recordings by sound
Saved in this browser, on this device. Nothing is uploaded.
Find every moment a phrase is spoken, by its sound.
- 1One recordingChoose one and play it.
- 2A selected soundDrag across a few words, half a second to two seconds.
- 3Matches in recordingsTierline compares that sound with every recording here.
- 4The source, in contextOpen each match at its moment and hear the audio around it.
Start with the examples
Try with example recordings
8 short excerpts of two public-domain recordings, 57 seconds of speech in all, already described and ready to search.
Ann Bernatitus · oral history, 1994
- I would rather go by submarineAnn Bernatitus · 8 s
- Their own minefieldAnn Bernatitus · 5 s
- Gotten the messageAnn Bernatitus · 5 s
- Taking us so long to get out thereAnn Bernatitus · 12 s
- Tea and chocolate cakeAnn Bernatitus · 6 s
- The slapping of the waterAnn Bernatitus · 8 s
Samuel Pepys · diary, June 1665
- Home at almost one in the morningNicole Lee (reading) · 7 s
- Not very late, to supper and to bedNicole Lee (reading) · 6 s
Hear the longer conversations they come from, with their people and records, in the Listening room, or take the thirty-second tour first.
Work with your own
Add recordings from my device
Choose or drop audio files: WAV, MP3, M4A (AAC), FLAC or Ogg, as your browser supports them. Up to 30 minutes each (15 on phones), 3 hours in total.
- You see each step as a file is read, fingerprinted, decoded and prepared for search.
- To keep a copy elsewhere, export a backup.
- Only add recordings you have permission to keep and listen to.
- 1Choose or play a recording
- 2Drag to select a short sound
- 3Find similar passages
- 4Open a match in context
What happens to your recordings
They stay on this device
Adding a file takes six steps, all in your browser: it is read, verified (its SHA-256 fingerprint, which also catches a file added twice), decoded, prepared for search, saved in the browser's own storage for this site, and then ready. The original is kept unchanged. Nothing is sent to Tierline or anyone else. Clearing this site's data in your browser deletes the workspace, so export a backup to keep a copy.
How the search works
When a recording is added, the browser describes its sound every 20 ms as 24 log-mel bands. A search slides your passage along every available recording with subsequence dynamic time warping, which allows the same words to be said a little faster or slower, and lists the closest passages. The distance shown is the average difference along that alignment: 0 means identical sound. It compares sound, not meaning, and works best when the same voice repeats a phrase. It is the browser matcher described on Explore; Tierline's XLS-R reference engine runs on a steward's own computer.
Access settings
Each recording can be available, held for review or withdrawn, and has its own settings for searching and for what a shared copy may contain: the audio, the description, the index notes and the speaker's name. This workspace applies those settings to its searches and exports. It is for one person in one browser, so it does not control who else can use this device. Tierline's local server checks every request against a collection's permissions; see Permissions and Local setup. Each of Tierline's archive pilots works under its own community's rules for who may search, hear, review and export: Partners & archives.
Limits
Up to 30 minutes and 300 MB per file (15 minutes on phones and small tablets), 3 hours of audio and 300 recordings in a workspace. Within these limits a search of the whole workspace finished in a few seconds in our measurements; slower devices take longer, and every search can be cancelled.