File processing time?

[SpinozaWeb]

Thanks for the new docker image. Installed - works) I have a laptop with intel i7-10875h and rtk 2070 max-q.
I converted the text in Russian. But even after 2.5 hours the conversion was not completed.
File size 800 kb
How long does the process take for you?

[DrewThomasson]

For me?

• Keep in mind the only way to get major speed ups is by having a CUDA capable Nvidia graphics card :/

The example book used for these test cases was the first book of guardians of ga'hoole

• which is a 5 hour audiobook

With a 4GB VRAM NVIDA(forgot the name)

• around real time reading speed
• around 5-6 hours

With a 12 gb VRAM Nvidia (3060?)

-A bit faster than real time reading speed
-around 4-3 hours

With only cpu on desktop (i9)?

• around slower than real time reading speed
• around 7-8 hours

With only CPU on laptop?

• holy crap it's so slow
• like 5 days

[ROBERT-MCDOWELL]

we really should find a way to help the user to install the right cuda since torch, xtts-v2 and python itself can install different default version which can be wrong.

[shadeau9]

having a title with ? char should n't be a problem. what's the title of the book?

Good to know, thank you! Looking into it further there was also a missing chapter wav file, which may have contributed. Regardless, I'm trying it again with some different parameters to see if it works. I don't want to derail this thread though so I'll try a few things on my end to see if I can figure it out.

Thanks for the quick responses!

[ROBERT-MCDOWELL]

no worries. I suggest to wait for the next git update to make your test. something like dozen of bugs/mistake/typo have been fixed...

[ricardocavalcante]

Why not Brazilian Portuguese? The population of Brazil is 10 times the Portugal...

[DrewThomasson]

Why not Brazilian Portuguese? The population of Brazil is 10 times the Portugal...

It looks like XTTS (The default model for ebook2audiobook) supports Brazilian Portuguese. If you select Portuguese and then give it a sample of someone speaking Brazilian Portuguese as the voice to clone then it should be able to say it in a Brazilian Portuguese accent.

ref - > coqui-ai/TTS#4127 (comment)

ALSO:

Please further discuss this in a different chat/issue or open a new one, as I don't think your question relates to this discussion at all...

[DrewThomasson]

If your really looking for fast on CPU tho, I have a repo that uses piper-tts instead (cause piper is also multi-lingual)

• But it's based on piper-tts (so no voice cloning you can only select from the pre-existing list)

-Unless you train your own model lol

• And it's also Siri quality voices

https://github.com/DrewThomasson/ebook2audiobookpiper-tts

• piper tts audio output file demo ⬇️

mini_story_long.-.Drew-2.mp4

• piper-tts huggingface demo space ⬇️

https://huggingface.co/spaces/drewThomasson/ebook2audiobookpiper-tts

[SyamsQ]

How to run GPU? I have rtx 3060

[shadeau9]

I wrote a guide for how to set this up with an Nvidia GPU in another discussion thread. I'd suggest following those instructions and commenting in that discussion thread.

[SyamsQ]

I have already installed Ebook2audiobook. Do I need to reinstall it after installing the Cuda Toolkit?

[suskozaver]

no, just run:

to check if u sucesfully installed the toolkit: docker run --rm -it --gpus=all nvcr.io/nvidia/k8s/cuda-sample:nbody nbody -gpu -benchmark

u should see something similar to this at the bottom:

Compute 8.6 CUDA device: [NVIDIA GeForce RTX 3080]
69632 bodies, total time for 10 iterations: 56.565 ms
= 857.180 billion interactions per second
= 17143.598 single-precision GFLOP/s at 20 flops per interaction

if you didn't get such results, perhaps just run: docker run --rm --gpus all nvidia/cuda:11.0-base nvidia-smi

after the tools were installed for sure, just run ebook2audiobook with:

docker run --pull always --rm --gpus all -p 7860:7860 athomasson2/ebook2audiobook

[PublicAB]

If the text is too long, an error will occur. Can this be resolved

Analyzing numbers, maths signs, dates and time to convert in words...
filter_chapter() error: expected string or buffer
Traceback (most recent call last):
File "/app/lib/functions.py", line 780, in filter_chapter
text = clock2words(text, lang, lang_iso1, tts_engine, is_num2words_compat)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/app/lib/functions.py", line 1227, in clock2words
return time_rx.sub(repl_num, text)
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/app/lib/functions.py", line 1217, in repl_num
phrase = lc["past"].format(hour=n2w(h), minute=n2w(mnt)) if mnt != 0 else lc["oclock"].format(hour=n2w(h))
^^^^^^
File "/app/lib/functions.py", line 1173, in n2w
word = math2words(n, lang, lang_iso1, tts_engine, is_num2words_compat)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/app/lib/functions.py", line 1252, in math2words
text = re.sub(r'(\d))', r'\1 : ', text)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/regex/regex.py", line 278, in sub
return pat.sub(repl, string, count, pos, endpos, concurrent, timeout)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: expected string or buffer
Caught DependencyError: filter_chapter() error: expected string or buffer
get_chapters() failed!

[ROBERT-MCDOWELL]

this has nothing to do with this discussion. open an issue, and provide the whole log from your command line until the except. provide also the file you want to convert if possible with the settings you used (which tts etc..)