Where do I begin?
 

I just received four .amr audio files from interviews that I am supposed to transcribe. After converting them to .wav, I found them to be of such poor quality that I cannot bring myself to do the work. I've spent some time applying various filters, but I have no no idea what I'm doing. I would be grateful for any advice you could offer as to how to proceed to improve the sound quality. I know so little about working with sound (terminology, which filters do what, etc.) that I can't even diagnose the problem!

AMR files are generally recorded by mobile devices, like cell phones. That doesn't leave much room for quality, given the crummy mic/speakers and tendency to compress the stuffing out of media files -- video, audio or photo.

I'll listen to it sometime today, and see what can be done. Will post back soon.

Have you had a chance to listen to it yet?
 

Have you had a chance to listen to the audioclip I attached yet? It's pretty bad. I'll be amazed if there is any hope for it, but hearing is believing. Maybe you're a miracle worker with sound.

There have been some recent issues on getting back to you quickly on this, (1) being vacation holidays, (2) winter illness and (3) a computer with issues. I'll be sure this gets analyzed and tested here soon for you.

Bad AMR Files
 

Thanks! I totally understand (especially since I had sinus surgery the week before Christmas). The arrival of these files was just bad all the way around!

I've had time to go over the audio file this morning, and was able to improve it a good bit.

What I heard was mostly shrill pitches, hiss noise, and deep male voice with grumble and muffling.

And these are the corrections that were applied, in this order, to get the audio sample in the attached MP3 file:

1. Restored in SoundForge 9, using the digitalFAQ Filter Preset "digitalFAQ Correct Harsh Bass Voices 1 (OTR)" -- that took out a lot of the muffling the male, although it added some slight shrill to the female (she is shrill regardless of what I do -- she just has one of those high-pitched female voices).
2. On the lats half, on the male voice, from "ok I think, according to my experience...", I opened Audacity, took a quick Noise Print with the "Noise Removal" effect, and then proceed to restore just that last section of the file, using mostly default settings, with the NR value set to "24" --- this removed most all of the hiss and background noise from the male segment.
3. Back in SoundForge, I used the "digitalFAQ Remove AC and FAN background noise" preset to reduce the female shrill some more, on just the female segment
4. Still in SoundForge, I created a new preset to further unmuffle the male voice, what I've named "digitalFAQ High Restore T - Hard Muffled Male Bass Fix (Test)", run over just the male segment (see image below)
5. Still in SF, I removed more shrill from the giggling girl, running a 30% normalize over the first segment of audio, which cut into its loudness
6. Finally, still SF, I ran a 70% normalize across the whole file

Attachment 581


And that created what you will hear in the attachment.

The highly-compressed source caused these issues, I generally tell people to just use a mini-cassette or full cassette recorder for audio. It may look antiquated, but I guarantee it tends to sound better! I use a Radio Shack full-sized recorder, and it's much better than most of those digital devices being sold now -- and much cheaper!

An MP3 was used because the WAV is too large, more than 8MB, and I didn't want to split-RAR it. It's been converted to 48khz stereo, too, to uncompress it for working with.

I still have a week or so left on my Sound Forge 30-day trial, and your suggestions will give me some direction in trying to fix the rest of the file. I'm still not totally convinced I am willing to transcribe this particular tape. I just don't know if I want to fight it the whole way, and my client will probably expect better results than I'll actually be able to produce. Thank you so much for helping me with this. I wouldn't have otherwise even bothered. (My client thanks you, too!)

Thanks again!

Well, it is an ugly recording. Not the worst I've heard -- probably not even in the 50% worst. It's just a compressed recording of a person with a voice that isn't necessarily conducive to being recorded. Neither the bassy male nor the squeaky girl are ideal speakers, and this fact is amplified exponentially on the inferior recording.

I think it could definitely be transcribed. I've done worse. Sometimes the accent of the male makes his words incoherent. I also find that those with English as a non-primary language use words wrong -- and that really throws you. Moreso when it's hard to hear AND the wrong word, as you can't guess at it.

I do wish you well on the project! :)