Word error rate: read a speech benchmark correctly
Word error rate counts editing differences against a reference transcript. It is useful for comparing a defined test, but it is not a guarantee for yo
resources
Recording and transcription resources
Practical guides with primary sources, worked examples and clear limitations.
Accuracy
Whisper Tiny, Base and Small in a browser
Choose a model by testing your recording on your device. Larger weights require more resources, and published scores do not establish your browser’s s
How to review an automatic transcript
Review the draft against the recording before you quote or rely on it. Prioritize mistakes that change meaning, then apply a clear editing convention
Planning
Transcription cost per minute: compare the whole job
Usage prices are only one part of transcription cost. Compare the service, currency, review work and billing rules before treating two rates as equiva
Speaking rates: plan with a rehearsal, not a myth
Use your own timed rehearsal to estimate a script’s duration. Published oral-reading rates are useful context, but they do not establish a universal p
How long transcription takes: estimate your actual work
Estimate transcription from a representative sample and your required review standard. Automatic processing time, first-run download time and human co
Recording
Audio formats for transcription: choose the right file
Use the original recording when your browser can decode it. Converting a file helps compatibility, but changing the extension or increasing quality se
Sample rates, mono audio and transcription memory
A speech model needs audio in its expected sample representation. Resampling and mixing channels prepare that input; they do not repair an unclear rec
Record clearer speech for a better transcript
Improve the source recording before relying on a larger model. Listen to a test, reduce avoidable noise and keep the microphone positioned for the per
Subtitles
Caption standards: WCAG, Netflix and BBC in context
Choose caption rules for the actual destination and audience. A line-length check helps, but accurate words, useful sound information and synchronized
SRT vs WebVTT: preserve text, timing and intent
Use the subtitle format required by your player or publishing platform. SRT and WebVTT both carry timed text, but they do not preserve exactly the sam
Privacy
How browser-local transcription handles your audio
Browser-local transcription brings the speech model to the recording. It avoids uploading audio for inference, while still requiring ordinary website