Voice Differentiation | Voters

Voice Differentiation

Alycia Straeter

As an adult female clinician, Heidi will somewhat frequently confuse my voice with my client's voice, especially if they are also an adult woman. While this is understandable, it would be great to be able to "code" in our voice as the clinician, so Heidi can better differentiate.
Note: 
I understand that Heidi is likely to pick up on more clinician-specific details in the note if the clinician is speaking consistently throughout the session, or if even brief personal narratives are shared. I will frequently start sessions with gripes about the weather, use short personal examples of implementing strategies with trying to cope or parent, speak more when I am working with younger clients to relate/connect, etc., so there are a number of situations in which Heidi will include things like "client was upset about the rainy weather and having to drive in it", "client reported succes with using a suggested strategy when working with their child", "client feels overstimulated with loud noises", and so on. Heidi being able to discern between clinician's voice (phrased as "therapist shared [for what purpose]") and client's voice (where phrasing would be consistent with "client was bothered by" type language) would be helpful for these instances.
Thanks in advance for any consideration!

Autopilot

Merged in a post:

Update Ai's ability to differentiate between speakers

Lamonte Gwynn

In many initial notes, the AI appears to attribute nearly all transcript content to the patient rather than distinguishing between the patient and the professional. When I ask it to reanalyze the transcript using context clues, it is sometimes able to correct some of these errors. However, it becomes much more accurate only when I explicitly identify who said what, as evidenced by its ability to provide rationale supporting those corrections.

This suggests the model is capable of making the distinction, but the initial note-generation process may be moving too quickly from transcript to output without enough transcript-level analysis. A valuable improvement would be for the AI to spend more time identifying speaker attribution before drafting the note, so the first-pass output is more accurate and requires less manual correction.

Autopilot

Merged in a post:

Voice specific recognition.

Graham Heyes

Maybe the Dictation mode could only respond to my voice, so that it doesn't pick up other voices in the office?

Graham Heyes

Yes it's that other conversations are being picked up and included in the Dictation even though the person speaking is female and I am male.

Heidi Team

Thanks for the suggestion — that would be really helpful in a busy office.
To make sure we capture this the right way for the team: 
1) Are you hoping for a “respond only to my voice” mode for dictation commands
 (e.g., punctuation/formatting), for the entire transcription
, or both?
2) Is the main issue other clinicians talking nearby, or patient/phone/background voices being picked up?
3) If you can share a short example (or a screenshot of your setup/mic type), what device/mic are you using on Desktop/Browser?
We’ll share this feedback with the team once we understand your use case a bit better.

Megan Menezes

Merged in a post:

Learn Dr's voice

Jason Lam

Teach the drs voice so it can recognise if the patient or dr spoke

Max Zoettl

I support this request. Zoom for example now allows uploading a voice sample of the account holder to improve differentiation between the account holder and others

Rio Krauss

Merged in a post:

Poor differentiation of who is talking

Sonia Shah

Mixes up what i have said with what patient (child, very different voice) and parent (very different voice) has said

Rio Krauss

Merged in a post:

Voice identification

Scott Beeby

Automatically detect when the practitioner is speaking Vs the patient (perhaps a voice identification steps with new account like setting up Siri on the iphone).

Sometime during sessions, lifestyle or personal factors are spoken about and Heidi records as if all events are the patients experience. The ability to omit personal info regarding the practitioner would be nice. Or for Heidi to determine when patient vs practitioner are talking about a therapeutic intervention or exercise to seperate advice and prescriptive information from patient experiences or current behaviours etc.

Rio Krauss

Merged in a post:

Can no longer differentiate voices

Claire McKellar

Over the last fortnight ( approx.) Heidi seems to have difficulty differentiating between my voice and the voice of my client.  Notes are completely mixed up stating the client is saying something which I actually said or even merging our two discussions and coming up with something very far from what actually was said.
In addition, Heidi doesn't seem to be capturing the entire session, either that or it is omitting information which is informative and critical therefore needing to be noted.
Previous to my recent experiences over the last approx. 2 weeks - I have had nothing but success and high praise for Heidi - I really wish to maintain my positive experience with Heidi - please help.

Anya Sharma

Merged in a post:

Dedicated "Half Scribing" - Capture only what the provider articulates

Andy Xuan

Having been at Heidi for a while now, I think I really like Heidi as a transcription / "half-scribing" tool.  A so-called cardinal disadvantage of virtual sessions over Heidi on the browser, whereby the mic is only listening to the provider - therefore non-provider voice is not even captured - is in fact a cardinal documentation advantage to some of us.  It allows the provider to "echo dictate" / summarize, and distill clinical information so that Heidi can transcribe better, provider can "think better", and patients can even "hear better" (as provider "rephrases" some of the verbiage, which, in all honesty, should be a provider's job and not necessarily "AI's job" in the first place). 
And this potentially streamlines consent process as well - since patient voice is not being recorded. 
I have taken such a personal liking to this workflow that I do this "half-scribing" exclusively during virtual sessions.  I simply tell patients that sometimes I may seem to "echo" or summarize what they say, both for my own brain to process and for my transcription software to document - at the same time.  And patients have the opportunity to hear and verify their input summarized by their provider, in real time.
Now since Speaker Diarization has been a feature in AI speech processing from very early on, I would ask Heidi to define this "half-scribing" workflow more formally, and extend it to in-person encounters.  
There is an added benefit in assuaging some patients who may harbour intense reservations about AI "recording their voices".

→