Label speakers in the transcript | Voters

Label speakers in the transcript

under review

Karin

It would be very helpful if Heidi could label speakers in the transcript ("Speaker 1", "Speaker 2" etc) as it is quite time consuming going through the transcript afterwards and having to figure out when I spoke vs when the patient spoke.

Autopilot

Merged in a post:

Speaker-identified transcripts with generic labels (Clinician/Patient)

Michael Hann

I would like to see full transcripts that differentiate between speakers using generic labels such as 'Clinician' and 'Patient' (rather than actual names). 
This would support de-identified batch processing of large volumes of clinical transcripts (e.g., 1000+ hours) for creating AI training data and digital professional clones. 
Current transcription services make it very difficult to de-identify large volumes through batch processing, so having built-in generic labels would be valuable. 
It would be very useful to be able to download full transcripts with generic speaker labels in large batches (e.g., a full month or year of transcripts), so they can be uploaded to AI tools for training digital professional clones.
Many thanks.

Liz from Heidi

Hi Michael Hann! Thanks for sharing this, having de-identified labels for speakers in the transcript makes sense, and I'll share this idea with our team to explore :)

Are there specific AI tools and training that you are hoping to use the transcript data for?

Michael Hann

Liz from Heidi Hello :) Yes. I want to create a digital professional clone using Delphi.ai (like I have seen some other clinicians doing extremely successfully recently).

Michael Hann

I have just got back from 10days in the US, where I saw just how exceptionally well some clinicians have got this working as an additional support for the clientele. I was extremely impressed with the quality of their AI clone (digital mind).

Autopilot

Merged in a post:

Transcription Recording Labels

Amy Webber

I'd like the ability to transcribe a patient visit, noting that start/stop time data as direct clinical care. Then restart the transcription after the patient visit, noting that start/stop time data is indirect clinical care.

Autopilot

Merged in a post:

Speaker Diarization for Psychotherapy Transcriptions

danielcovolomazzo@gmail.com

Hi Heidi team,
I'm a clinical psychologist and I use Heidi daily for transcribing psychotherapy sessions. Overall, the tool is excellent — stable, accurate, and very well designed.
One feature that would make a significant difference for mental health professionals is speaker diarization (i.e., separating therapist speech from patient speech in the transcript). In psychotherapy, being able to clearly distinguish who said what is essential for clinical supervision, case formulation, and self-evaluation of therapeutic technique.
Without diarization, the raw transcript is still useful, but it requires considerable extra effort to identify speaker turns — especially in moments of brief exchanges or when both speakers use similar phrasing. For clinicians who review sessions for quality improvement, training, or peer supervision, having labeled speaker turns would be transformative.
This would also open up possibilities for automated metrics (e.g., therapist talk-time ratio, turn-taking patterns) that are increasingly valued in evidence-based psychotherapy research and training.
Thank you for building such a great product. Adding diarization would truly make Heidi the gold standard for mental health documentation.
Best regards

Liz from Heidi

Hi Amy Webber, thanks for sharing this feedback! We're rolling out an update to our transcription page that will show the time when you start, stop, and resume a transcription (see image below). When you resume the transcription for indirect clinical care, you can just state that so Heidi captures that in the transcript now. 
E.g. after the patient leaves, click "Resume" and say something along the lines of "Beginning indirect clinical care" before transcribing the rest of your notes. 
Hope that helps :)

Autopilot

Merged in a post:

Transcript should differentiate different speakers with titles

Greg Curnew

It would be amazing if you could differentiate between the different speakers, such as the clinician, the patient, speaker 1, speaker 2 and other speakers within the transcript, so we can follow who has said what.

Tomasz Markiewicz

It would be also a way to avoid errors as noting doctor's part as coming from a patient. It has happened a lot.

Renata Chalfin

You could have a button that allows you to automatically change the formatting of the transcript format to include Labelled speakers. Heidi doesn't necessarily have to take it as fact - you could label it after outputting it to the user instead of before generating the note? But this would be super helpful because so many times patients self-diagnose during my examination of them and Heidi treats

that

as fact and includes it as a diagnosis or history in the note even though I disagreed...

Tom

Just for some context on this one, the reason we don't do it today is because the outputs are consistently better if we don't separate them, because any error in the separation/diarisation of speakers is fully trusted by Heidi and leads to worse outputs very often. Basically the convenience of the speaker labels is outweighed by the rate of error in labelling correctly. But we understand the need, we're investigating ways to do it that don't regress the note quality, thanks for all your comments.

Ladan shaikh

I agree with most comments. It is crucial for there to be a differentiation with what the patient reports versus our interpretations for the patient and not to confound the two as one is an offering for the client and the other is their experience of the events. In terms of psychotherapy it becomes confusing if these two are not differentiated.

→

Photo Viewer

Photo Viewer