Forked from srvk/srvk-eesen-offline-transcriber: contains changes to decode with Tensorflow models (as found on PSC Bridges cluster environment) - still requires GPU - needs to be refined to run without depending on e.g. libcuda.so.1
srvk/eesen customized version of Tanel Alumae's kaldi-offline-transcriber
You probably want to use this inside the SRVK's Eesen Transcriber, not on its own.
speech2text.sh
- Transcribe audio/video file and produce several output formats at once (plaintext, subtitles, NIST CTM scoring input, Audacity labels)vids2web.sh
- Transcribe and create video subtitles and searchable index in a web pagerun-segmented.sh
- If you have your own segmentation file this may improve transcription accuracyrun-scored.sh
- If you have STM ground truth as well as audio/video, produce NIST SCLITE scoring results inbuild/trans/<videoname>/eesen/decode/score_*
run-scored-8k.sh
- Same but for 8khz audio such as Switchboard corpusbatch.sh
- Queue several files for transcriptionslurm.sh
- for batch processing, edit to change which transcribe script is used (speech2text.sh by default)mkpages.sh
- Make/update web pages from video and transcription outputwatch.sh
- Run this to start watching a shared folder for files to be transcribedpath.sh
- set up the PATH environment variable for the aboveMakefile
- master control for transcriber