Lana

no recogniser loading Praat…

Server-side providers Direct endpoint

Providers configured on the machine serving this page

provider

Providers come from providers.json beside serve.py. API keys stay on that machine: this page is told only whether a key is present. Requests are proxied, so there is no CORS to configure and no key in the browser. Copy providers.example.json to start.

Talk to one endpoint directly from the browser

base URL

not checked

control path

Anything exposing POST {base}/v1/audio/transcriptions in the OpenAI shape. A cross-origin endpoint must send Access-Control-Allow-Origin, which is the usual reason this mode fails where the proxied one works. If the endpoint also serves {control}/progress, {control}/partial and {control}/cancel, you get live progress, tiers that fill while the job runs, and a working Stop button.

Signal no file drag to set the analysis window · click to seek

—

Spectrogram f0 F1 F2 F3 f0 formants

time—

cursor Hz—

f0—

F1—

F2—

F3—

intensity—

label—

Tiers no annotation yet

double-click a tier to type a label · drag a boundary to move it

name tiers by model

Transcript fills as the recogniser reports

nothing yet

Recognition one tier per model

Word tier no model

orthographic transcription

Phone tier no model

IPA read off the acoustics, with CTC spike times

whisper size

Choosing the same model for both tiers runs it once.

Analysis Praat arguments, as typed

Spectrogram

window length (s) max frequency (Hz) time step (s) frequency step (Hz) window shape dynamic range (dB)

Pitch (autocorrelation)

time step (s) floor (Hz) ceiling (Hz)

Formants (Burg)

time step (s) max number max formant (Hz) window length (s) pre-emphasis from (Hz)

Segments —

#	label	voice	start	end	dur ms