We operate a specialized service for improving sound quality, not voice analysis.
There are many requests for noise removal from the general public, but we have a consultation about "voice analysis" about once a month. This is not basically handled.
The only most recent voice analysis
The only case I received most recently was "I want you to check whether the conversation content of the voice data has been edited." While checking the audio material, I checked for any editing scars in the areas of concern.
It is relatively easy to judge that the one edited by an amateur has a blatant joint. However, when edited by a professional voice specialist, the seams and background noise are completely composited, so even I, a professional, can hardly tell.
I couldn't say that the sound source I received had a clear edit.
The client is somewhat reluctant to say "I didn't have it" after receiving a guarantee for the need to "find an edit point".
For that reason, we basically do not accept it.
There are a certain number of consultations from people with hallucinations, but we do not accept them.
I have hallucinations due to mental illness or similar symptoms, and I "always hear bad words", so I recorded it on the spot, but there are a certain number of requests to check if it is included, but I have received it. not.
It seems that the faint sound in the room may sound like that. Of course, we do not deny the disease or those who are suspected of it.
In the case of consultation using the above as an example, in most cases you cannot hear anything even if you actually check the data, increase the volume by several tens of times and cut the noise. (Although we basically handle data confirmation at the time of contact) We cannot accept work that has no room for improvement.
Because we are launching a free trial as a business, if we cannot improve 100%, we will have a deficit for the work.
It is the same as it is impossible to stretch low resolution photo data
For voice restoration, iZotope's RX8 Advanced (2020 latest version) application is mainly used for business.
I'm hitting.
Since the sound component called the spectrogram meter (the image at the top of the blog) can be graphicized, you can roughly grasp what kind of component is mixed in the state of the sound.
However, very small sounds that can be heard or cannot be heard are not converted into data in low-resolution sound sources. Also, because environmental sounds are a mixture of various sounds, in most cases it is not possible to find a specific component in the meter.
Sound has similar characteristics to image data. Even if you stretch a low-resolution photographic image on your computer, it will look like a mosaic, and you may have some experience that you don't understand.
Sound quality improvement is basically to make the sounds that are heard easier to hear.
If you can't hear the target voice even faintly in the audio data, the improvement probability is zero. If you can't see a word in your own recorded voice that tells you what you're saying, even if it's fragmentary, you probably can't clarify the content.
At best, a smartphone voice memo or an inexpensive IC recorder can record about 50 cm from the microphone clearly. Attempts to capture evidence of an affair in one corner of the room often fail.
Of course, with a recorder (linear PCM recorder) with a high resolution and a built-in high-performance microphone, the improvement probability will increase a little more.
I can hear it by myself, but if you want to find out if you can hear it, I recommend that you first ask your family and other close relatives to listen to the recorded data, saying, "I think I can hear something."
Summary
Voice analysis is basically not accepted
If you can't hear the conversation at all when you ask yourself, the improvement probability is zero.
Services that improve the sounds you hear
Making a small sound louder is similar to stretching coarse photo data
We are improving more than 100 hard-to-listen sound sources a year.
Please contact us if you have any problems with the sound source.