Audio transcriptions

[RoosRoos 1]

I had to cut down audio files to below 14MB to make audio transcriptions work. The feature is wonderfull however this is a cumbersome activity and strange that it says files up to 25MB are possible. Will this be addressed, workable at least for 25MB and preferably higher? I did notice some glitches on the Dutch language however it saves time and concentration to not have to transcribe myself.

[gazumped 12,226]

Hi - If you haven't already,  please send your comment to feedback@evernote.com - we're mostly other users here.  Have you tried using lower quality audio to reduce file size?  What device(s) are you using to record audio,  or where are you sourcing the files?

[RoosRoos 1]

HI, I will send it to evernote - good idea. It was a zoom recording and I downsized it from mp4 to mp3 and then I had to cut it in pieces. The zoom recording was made by somebody else so don't know what device was used.

 

[gazumped 12,226]

Downsizing from MP4 to MP3 should allow you to reduce the quality of a file without losing content,  though in extreme cases I guess the transcription might suffer if the quality is too low.  It would be good if Evernote published some guidelines on what parameters are required for transcription - I asked AI for a list.  Maybe Evernote could comment on the minimums required for some of these factors.

• Bit rate: Measures the amount of data processed per unit of time, usually expressed in kilobits per second (kbps). Higher bit rates generally mean better quality.
• Sample rate: The number of samples of audio taken per second, measured in Hertz (Hz) or kilohertz (kHz). Common rates are 44.1 kHz and 48 kHz.
• Bit depth: The number of bits used to represent each sample. Common depths are 16-bit and 24-bit. Higher bit depth allows for more dynamic range.
• File format: Such as WAV, MP3, FLAC, AAC, etc. Some formats are lossless while others are lossy.
• Dynamic range: The difference between the loudest and quietest parts of the recording.
• Signal-to-noise ratio (SNR): The level of desired signal compared to background noise.
• Frequency response: The range of frequencies the recording can reproduce accurately.

Then,  if you're using a third party file source,  or recording via an external app,  you could minimise the output file size with confidence.  (I sent this list to feedback@ as a suggestion.)

[PinkElephant 9,015]

Language spoken, and dialects that are - sorry - out of luck … 🤷‍♂️

[l0nEr 8]

Hi @gazumped

Did you get the requirements from evernote? 

Ive been trying with various mp3 files, all below 12mb and about 1 hour long... all didnt work. not sure why.

Thanks.

 

[Jon/t 1,753]

2 hours ago, l0nEr said:

Hi @gazumped

Did you get the requirements from evernote? 

Ive been trying with various mp3 files, all below 12mb and about 1 hour long... all didnt work. not sure why.

Thanks.

 

Are you sure it's a proper mp3 as 12MB seems a bit small for an hour. 

Spoken word exported at 128kps usually works out at around 1MB per minute ISH. 

The limit for Evernote is 25mb so around 25 mins. 

[l0nEr 8]

7 hours ago, Jon/t said:

Are you sure it's a proper mp3 as 12MB seems a bit small for an hour. 

Spoken word exported at 128kps usually works out at around 1MB per minute ISH. 

The limit for Evernote is 25mb so around 25 mins. 

i used audacity to shrink downsize the file to about variable bitrate around 32kbps
sample rate i did 16 khz
tried to make it mono

i had no issues transcribing that file on whisper v3 (run locally) though or assembly.ai

https://dev.to/mxro/optimise-openai-whisper-api-audio-format-sampling-rate-and-quality-29fj?comments_sort=top 

[Jon/t 1,753]

11 minutes ago, l0nEr said:

i used audacity to shrink downsize the file to about variable bitrate around 32kbps
sample rate i did 16 khz
tried to make it mono

i had no issues transcribing that file on whisper v3 (run locally) though or assembly.ai

https://dev.to/mxro/optimise-openai-whisper-api-audio-format-sampling-rate-and-quality-29fj?comments_sort=top 

That's very odd as I think, but I may be wrong, they use whisper ai for the transcription!

Send in a support ticket and reply to the auto email with your logs (help menu) and if you can, a copy of the audio.

[gazumped 12,226]

12 hours ago, l0nEr said:

Did you get the requirements from evernote? 

Hi.  No - I used feedback to suggest the idea,  but that doesn't prompt them to comment.  As this is a third-party service,  getting some 'official' feedback might take a while.

Let us know what sort of a reaction you get to a support request!

[EddieO23 0]

You guys arent the only one having transcribing issues. im trying to transcribe .mp4 files under 25mb and its not even showing up. on desktop or mobile.