icarus Sigil plugin

[DiapDealer]

Don't mean to butt in ... just wanted to say that many times, memory errors involving large file parsing/processing and Python can often be remedied by making sure you're using a 64-bit version of Python (and have tons of RAM if possible). You may be already, but I just wanted to mention it.

[AlPe]

Originally Posted by eclipse123

I tried a few audio books - each one is one very long file (6-12 hours) and they all get the same problem, when I try shorter audio files, they work fine.

That explains the issue.

Yes, at the moment aeneas is limited by the amount of RAM your machine has. From https://github.com/readbeyond/aeneas...ssing-features :

No protection against memory trashing if you feed extremely long audio files

On my 4 GB RAM laptop, the longest single file I was able to run was about ~1h30m --- the exact size depends on your OS consumption of RAM, other programs running concurrently, etc. (Also on aeneasweb.org there are 4 GB of RAM.)

So, the only thing you can do at the moment consists in splitting that audio file into several pieces (chapters?), and create a task (= audio/text pair) for each.

In theory this memory issue can be worked around --- instead of loading the entire WAVE file in memory, the MFCCs can be computed in "stream mode", and the DTW computed piece-wise --- but that would require a substantial amount of work that I cannot afford to do now.

Thanks for reporting this problem, I will clarify that in the docs, and add a check in aeneas to trigger a human-readable exception.

EDIT: logged as a new issue: https://github.com/readbeyond/aeneas/issues/57

[eclipse123]

Thanks DiapDealer, AlPe,

I a have 8GB RAM, and windows usually knows to handle more then that if needed. However - it is not limitless - so memory management in Aeneas can definitely be the cause of this issues.

So - next problem :-):
I tried a different book, with smaller files per chapter. and this time I passed the Memory Hurdle and was able to progress. It started processing the first audio file properly, but crashed in segment 273: see the error below.

Any idea?

Another question - what happens if the audio file have slightly different text then the xhtml file? will it cause the software to exit? (I don't think this is the case, but just wondering)


[DEBU] ESPEAKWrapper: Calling with arguments 'espeak -v en -w c:\users\eyal\appdata\local\temp\tmpnph_hj.wav'
[DEBU] ESPEAKWrapper: Calling with text 'The small king suddenly smiled. Γא£Truth be told, I stopped thinking of it as a curse years ago, lad.Γא¥ '
[DEBU] ESPEAKWrapper: Calling espeak ... done
[DEBU] AudioFileMonoWAVE: Loading audio data...
[DEBU] AudioFileMonoWAVE: Sample length: 6.717914
[DEBU] AudioFileMonoWAVE: Sample rate: 22050.000000
[DEBU] AudioFileMonoWAVE: Audio format: pcm16
[DEBU] AudioFileMonoWAVE: Loading audio data... done
[DEBU] ESPEAKWrapper: Duration of 'c:\users\eyal\appdata\local\temp\tmpnph_hj.wav': 6.717914
[DEBU] ESPEAKWrapper: Synthesizing using pure Python... done
[DEBU] ESPEAKWrapper: Removing temporary file 'c:\users\eyal\appdata\local\temp\tmpnph_hj.wav'
[DEBU] ESPEAKWrapper: Synthesizing text... done
[DEBU] ESPEAKWrapper: Fragment 273 starts at: 2375.777098
[DEBU] ESPEAKWrapper: Fragment 273 duration: 6.717914
[DEBU] AudioFileMonoWAVE: Appending audio data...
[DEBU] AudioFileMonoWAVE: audio data is not None: returning True
[DEBU] ESPEAKWrapper: Synthesizing using pure Python... failed
[WARN] ESPEAKWrapper: An unexpected exception occurred while running pure Python code:
[WARN] ESPEAKWrapper:
[CRIT] ExecuteTask: STEP 3 FAILURE
[DEBU] ExecuteTask: Cleaning up...
[DEBU] ExecuteTask: Removing file 'c:\users\eyal\appdata\local\temp\tmpk2ro99.wav'
[DEBU] ExecuteTask: Cleaning up... done
[CRIT] ExecuteJob: Error while executing task 't000001.xhtml': The given log message is not a Unicode string
[CRIT] CLI: An unexpected Exception occurred while executing the job:
[ERRO] An unexpected Exception occurred while executing the job:
[CRIT] CLI: Error while executing task 't000001.xhtml': The given log message is not a Unicode string
[ERRO] Error while executing task 't000001.xhtml': The given log message is not a Unicode string
[DEBU] CLI: Execution completed with code 1

[AlPe]

Since you are running Windows, espeak is called via subprocess (since the cew C extension is not available), generating a temporary WAVE file for each fragment. Then, the contents of the latter are read and appended to create the global WAVE file.

From the log, it seems that the "append" operation failed, I cannot tell why from the log. It should not be a memory issue, as I see that your audio is below 1h of duration.

This string looks suspicious:

'The small king suddenly smiled. Γא£Truth be told, I stopped thinking of it as a curse years ago, lad.Γא¥ '

as it seems to contain spurious characters ("Γא£", "Γא¥") --- albeit I cannot think a reason why it should make aeneas fail. In fact, from the log it seems that espeak synthesized it correctly. Anyway, you might want to clean that text and see if it helps.

===

As I wrote, Windows at the moment is not officially supported by aeneas, mainly because I am the only developer and I do not use Windows.

If you want to run aeneas on your Windows machine, I strongly recommend using VirtualBox + Vagrant: https://github.com/readbeyond/aeneas-vagrant

[AlPe]

Originally Posted by eclipse123

Another question - what happens if the audio file have slightly different text then the xhtml file? will it cause the software to exit? (I don't think this is the case, but just wondering)

aeneas, doing forced alignment via MFCC+DTW and not doing speech recognition, is "robust against misspelled/mispronounced words, local rearrangements of words, background noise/sporadic spikes". (From: https://github.com/readbeyond/aeneas...orted-features )

[eclipse123]

Thanks AlPe,
yes - I think the next step is to take this to linux...
I'll try that in a month or so when I have a bit more time. thanks a lot for your help and pointers so far.

BTW, I did try a few ebook readers with some sample books from your site, and as you noted, none of them support this format :-(

Thanks
Eyal

[AlPe]

You can install VirtualBox, create a Linux (Debian or Ubuntu) virtual machine, and run aeneas there.

aeneas-vagrant automates the setup, you just need to have VirtualBox and Vagrant installed.

=== === ===

BTW, I plan to release the next version of aeneas by the end of February 2016. I will rewrite the C extension that computes the MFCCs so that they will be computed in "stream mode", without requiring the whole WAVE file to be load in memory. This way aeneas should be able to process huge files.

(Note: in theory, for RIFF WAVE PCM 16bit 16 kHz --- the minimum setting for achieving good results --- the audio file cannot be longer than ~37.28 hours.)

=== === ===

Reading systems: yes, the app landscape is sad and disappointing, especially considering the fact that e.g. the DAISY Consortium voted for (reflowable) EPUB 3 + Media Overlays as the evolution of their DAISY Talking Book format.

[AlPe]

Added icarus 0.0.3. Please use Sigil 0.9.3 and ePub3-itizer 0.3.6 with it. (Otherwise stick with Sigil 0.9.2 and ePub3-itizer 0.3.5 and icarus 0.0.2)

Also note that today I released a new version of aeneas (1.4.1) and updated the http://aeneasweb.org Web App (0.0.8).

[AlPe]

Is someone still using the icarus plugin?

[AlPe]

In case someone is still using icarus, today I tested that the following triple works:

- Sigil 0.9.7
- icarus 0.0.3
- ePub3-itizer 0.3.6

Enjoy

[olavimmanuel]

Thank you AlPe for the amazing Icarus and aeneas!

For anyone else using icarus; the following (latest versions atm) triple works:

- Sigil 0.9.8
- icarus 0.0.3
- ePub3-itizer 0.3.8

Enjoy

[KevinH]

Glad to hear it!