Simple and tranparent
When needed, I will be open about my code and documentation. Together we can make sure both the process and the transcription meet your standards. I'm able to provide a VOG/SOC and will sign NDA agreements if necessary.
State of the art models
The latest opensource models are incorporated in my program. This ensures both the best possible performance and safety. The used models perform excellent for multiple languages. Further processing like extractive summaries are also possible.
Possibility to work offline
As a result of the local architecture, the program can be operated without a functioning network connection.
Speaker identification
Using an opensource speaker-diarization model , my program is able to distinguish the different speakers in the recording. The best performance is achieved when the number of speakers is defined beforehand. During the process we keep in touch to ensure all speakers are given their appropriate names and titles in the final product.
Transcription
After the speakers have been differentiated in the audiofile, the transcription will be made. OpenAI's opensource model Whisper ensures the best possible quality. OpenAI reports a WER of 4.1-9.3% for English and 4.3-5.2% for Dutch. Of course this is heavily dependent on the quality of the input.
A filter on very short fragments is applied in an effort to weed out all filler words. This results in a very clean an readable final report.
Presentation
Forget a boring text document. Conversations are most comfertable read as a chat and will be formatted as one as well. See an example here. The colors on the text boxes will be given custom colors, so all parties and roles in recording will be easily identifiable. Of course it is possible for the colors to match the style of your company. It is always possible to get the transcriptions in a .doc or .pdf format as well.
