For language programs
Working with a language program
What a pilot involves: what Tierline does with your recordings, what each side provides, what it costs, and how either side can stop. Written for a program director without technical staff.
In one paragraph
What Tierline does for your program
You have recordings. Someone needs to find every place a particular word or phrase occurs in them. Right now that means listening. Tierline lets that person highlight one example and get back a ranked list of likely matches across the collection, running on a computer in your building, under access rules you write and can change or revoke at any time.
It does not transcribe, and it does not decide what anything means. It narrows what a person has to listen to, and keeps every result tied to its recording and its permissions.
Process
How a pilot would run
Listen
Begin with your actual workflow and backlog. We do not arrive assuming Tierline is the answer.
Scope
Choose one bounded problem: one collection, one search task, one success criterion.
Agree
Write down access rules, data handling, cultural protocol, publication permissions, compensation expectations, security requirements and exit terms — before any restricted material changes hands.
Pilot
A deliberately narrow deployment: one collection, one policy, one operator, end to end.
Evaluate
Measure whether it reduced review effort and whether language workers actually found the results useful.
Decide
You decide whether to expand, revise, pause or end the work.
Terms
What each side brings
What we would need
- Audio in whatever format you have it.
- One person we can ask questions of.
- Your access rules, expressed as who may search, play, annotate, export, train on, or publish the material.
- Later, only if useful, a spelling-to-sound rule set you control.
What you would get
- Software that runs on a local machine, with no network connection needed after setup.
- Search by highlighted example, needing no transcription work from you.
- A hash-chained log of successful persisted searches, reviews, exports, policy changes and core processing operations.
- Exports into ELAN and Praat, feeding the tools you already use.
What it costs
- Nothing for the software. Apache-2.0, permanently.
- Hardware at cost, or supplied by us if funding exists.
- No subscription, no per-hour fee, no support contract.
What we do not ask
- Permission to publish research is not a condition of use.
- Your recordings are not training data unless a policy you control explicitly grants it.
- You are not required to be a case study, and not required to be named.
Good to know
Before you get in touch
- Tierline is in use in 4 archive pilots under signed community agreements, and 2 official partnerships are in co-design (see Partners & archives). A new collaboration starts small, on terms your program sets.
- Retrieval has been evaluated on held-out speech: a correct passage in the first five results for 83.29% of 826 positive queries, and weaker on damaged audio (see Evaluation). Results on your recordings will differ, and are measured with you rather than assumed.
- It does not transcribe. If transcription is what you need, a speech-recognition tool is the better fit; Tierline helps where a transcript does not exist or cannot be trusted.
- It is built by a small team, with community stewards and reviewers, so a new pilot starts when there is time to support it well.
- Tierline is supported by donations, sponsorships and in-kind help (see Funding & stewardship). Nothing is sought from you.