[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [AUDITORY] converting masking thresholds to masker levels of speech sounds

If I may (as the author of Psychoacoustic Model 2) stick a word in here.

1) if you know what is voiced and unvoiced, use 15.5 dB SNR required for voiced and 5.5dB for unvoiced SNR rather than do all the tonality calculations. Pitch strength will probably carry you through, here. (0 -> no pitch) 1-> all pitch, no noise) if you have a representation of that.

2) I would shorten the stride of the calculation and do it on a millisecond by millisecond basis.

3) Proprietary issues prevent me explaining how to do this a lot better, sorry!

jj


On Wed, Jan 29, 2020 at 9:17 PM Frederico Pereira <pereira.frederico@xxxxxxxxx> wrote:
Hi Mengli,
No I haven´t tried other models nor did I account for temporal models at this point. Having temporal effects integrated in the psycouacoustic model would be something quite unique from what I´ve seen from existing routines yes! 


regards,

Frederico

On Wed, Jan 29, 2020 at 1:08 PM Feng, Mengli (2018) <Mengli.Feng.2018@xxxxxxxxxxxxxxx> wrote:
Hi Frederico,

Thanks very much for the code!

I did the same thing using ISO psychoacoustic model 2. Was thinking about using models to account for temporal effect. Have you tried more advanced auditory models?

Best wishes,
Mengli

-- 
Mengli Feng
PhD Student
PGR Collective EPMS School Convenor
 
Audio, Biosignals and Machine Learning Group
Department of Electronic Engineering
Royal Holloway, University of London 
 
Research Interest:
Speech/ voice production and perception
Ongoing Project:
the perceptual effect of Bone-conducted sound of own voice
>>> Pure Page

From: Frederico Pereira <pereira.frederico@xxxxxxxxx>
Sent: Wednesday, January 29, 2020 12:15 pm
To: Feng, Mengli (2018)
Cc: AUDITORY@xxxxxxxxxxxxxxx
Subject: Re: [AUDITORY] converting masking thresholds to masker levels of speech sounds
 
Hi Mengli,

I´m currently working on something similar and I´ve been developing on top of the code and psychoacoustic models based on:
ISO/IEC 11172-3:1993, Information technology – Coding of moving pictures and associated audio for digital storage media at up to about 1,5 Mbit/s – Part 3: Audio

Hoping this is of some help to you.

regards,

Frederico

On Tue, Jan 28, 2020 at 5:19 AM Feng, Mengli (2018) <Mengli.Feng.2018@xxxxxxxxxxxxxxx> wrote:
Dear All,
 
I am trying to convert masking curves into the frequency responses of the original maskers (single speech sounds). The maskees I am using are narrow band noises at different frequencies.
 
It has taken me enormous effort to find an auditory model to make accurate predictions, considering the maskers are complex tones with multiple harmonics in high frequency region. Might anyone provide some guidance or advice on finding a suitable model? 
 
Is it even possible to do such prediction knowing only the frequency responses of the maskees and the masking thresholds given that temporal effects would inevitably appear because of the higher harmonics in human speech sounds? Any opinions?
 
Any suggestion would be greatly appreciated!
 
Best Regards,
Mengli
 

-- 
Mengli Feng
PhD Student
PGR Collective EPMS School Convenor
 
Audio, Biosignals and Machine Learning Group
Department of Electronic Engineering
Royal Holloway, University of London 
 
Research Interest:
Speech/ voice production and perception
Ongoing Project:
the perceptual effect of Bone-conducted sound of own voice
>>> Pure Page



--
Frederico Pereira
Mobile:+351937356301
Email:pereira.frederico@xxxxxxxxx


--
Frederico Pereira
Mobile:+61409066693
Email:pereira.frederico@xxxxxxxxx


--
James D. (jj) Johnston
Independent Audio and Electroacoustics Consultant