What happened?
I'm trying the sdk on a Next.js application. I wanted to use the prosody functionalities, and, although the monotony detections seems to work well, the breaks (missing and unexpected) seems really broken, by having a confidence score consistently around 1 for missing breaks (and sometimes, surpassing it) and detections regarding unexpected breaks passing 1 when detected. Plus, it seems to detect just breaks without regarding the context, for example, pauses at the end of a sentence are detected as unexpected breaks. I'm handling the audio both at wav and compressed format.
with various configurations:
- continuous assessment
- Recognize once assessment
- wav browser recording
- compressed browser webm casted as wav
- Even with a local downloaded .wav file
- Both scripted and unscripted
I'm getting the same results along the versions from 1.44 to the 1.46.
These are visualizers of the result:
With missingBreaks highlighted
Without missingBreaks highlighted
Regarding the response, the unexpected break detection is something like this
"Prosody": { "Break": { "ErrorTypes": [ "None" ], "UnexpectedBreak": { "Confidence": 1.3719347 }, "MissingBreak": { "Confidence": 0 }, "BreakLength": 3800000 }, "Intonation": { "ErrorTypes": [], "Monotone": { "SyllablePitchDeltaConfidence": 0.85779065 } } }
While the numerous missingBreak cases follows something like this (while also surpassing the confidence 1!)
"Prosody": { "Break": { "ErrorTypes": [ "None" ], "UnexpectedBreak": { "Confidence": 3.6103543e-8 }, "MissingBreak": { "Confidence": 1 }, "BreakLength": 0 }, "Intonation": { "ErrorTypes": [], "Monotone": { "SyllablePitchDeltaConfidence": 0.85779065 } } }
What could be the reason of this? I've tried numerous configurations from my side but the results i'm getting are all the same, the Offset and Duration values all make complete sense regarding the real duration of the audio, i've also tried different framerates of the audio.
Let me know if you need details of the code, I don't want to make the issue message too chunky. Thanks
Version
1.36.0 (Latest)
What browser/platform are you seeing the problem on?
Node
Relevant log output
What happened?
I'm trying the sdk on a Next.js application. I wanted to use the prosody functionalities, and, although the monotony detections seems to work well, the breaks (missing and unexpected) seems really broken, by having a confidence score consistently around 1 for missing breaks (and sometimes, surpassing it) and detections regarding unexpected breaks passing 1 when detected. Plus, it seems to detect just breaks without regarding the context, for example, pauses at the end of a sentence are detected as unexpected breaks. I'm handling the audio both at wav and compressed format.
with various configurations:
I'm getting the same results along the versions from 1.44 to the 1.46.
These are visualizers of the result:
With missingBreaks highlighted
Without missingBreaks highlighted
Regarding the response, the unexpected break detection is something like this
"Prosody": { "Break": { "ErrorTypes": [ "None" ], "UnexpectedBreak": { "Confidence": 1.3719347 }, "MissingBreak": { "Confidence": 0 }, "BreakLength": 3800000 }, "Intonation": { "ErrorTypes": [], "Monotone": { "SyllablePitchDeltaConfidence": 0.85779065 } } }While the numerous missingBreak cases follows something like this (while also surpassing the confidence 1!)
"Prosody": { "Break": { "ErrorTypes": [ "None" ], "UnexpectedBreak": { "Confidence": 3.6103543e-8 }, "MissingBreak": { "Confidence": 1 }, "BreakLength": 0 }, "Intonation": { "ErrorTypes": [], "Monotone": { "SyllablePitchDeltaConfidence": 0.85779065 } } }What could be the reason of this? I've tried numerous configurations from my side but the results i'm getting are all the same, the Offset and Duration values all make complete sense regarding the real duration of the audio, i've also tried different framerates of the audio.
Let me know if you need details of the code, I don't want to make the issue message too chunky. Thanks
Version
1.36.0 (Latest)
What browser/platform are you seeing the problem on?
Node
Relevant log output