README.md

Sentence Mixing generator - command line interface

This program uses the Sentence Mixing library to generate sentence mixing styled Youtube Poops.

You can see plenty of examples of application results here (French language)

You can get it via the release tab or install it manually.

Installation

Here is a tutorial video showing both installation and usage of the project (French version).

Release

1. Choose a release version
2. Uncompress the release archive

Language support

The default language of releases is French. You can however use another language among the ones proposed in aligner/SM-Dictionaries.

All you have to do is to download the appropriate MFA pretrained model and to update config.json in the release's root directory.

Example (English):

1. Download the English accoustic model and put it in aligner/pretrained_models
2. In config.json, change:
   1. lang value to en
   2. trained_model value to aligner/pretrained_models/english.zip
   3. dict_path value to aligner/SM-Dictionaries/en/english.dict
   4. dict_consonant_vowel_path value to aligner/SM-Dictionaries/en/en_consonant_vowel.dict

(For Windows release, replace / characters by \\ characters).

From source

1. Install requirements with pip install -r requirements.txt
2. Set up the external dependencies by following the tutorial in SM library github page

Sanity check

Use the file sample_cli.py, a minimal example for the sentence mixing library. If no error pops, the Sentence Mixing library and the external dependencies are correctly installed.

Example:
python sample_cli.py "Bonjour" ../config.json https://www.youtube.com/watch?v=4czmCgJryUM

Program usage

Here is a tutorial video showing both installation and usage of the project (French version). The following sections will explain you how to run and interact with the CLI Youtube Poop generation program.

Running the script

The file to run may vary:

• Linux release: SentenceMixerCLI
• Windows release: SentenceMixerCLI.exe
• From source: video_creator_main.py

Run the executable with your json configuration file as first argument, and as many Youtube URLs as you want.

Examples:

• Linux: ./SentenceMixerCLI config.json 'https://www.youtube.com/watch?v=udmDOaHN9no' 'https://www.youtube.com/watch?v=7CTH59GCNko'
• Windows: .\SentenceMixerCLI.exe config.json 'https://www.youtube.com/watch?v=udmDOaHN9no' 'https://www.youtube.com/watch?v=7CTH59GCNko'
• From source: python video_creator_main.py config.json 'https://www.youtube.com/watch?v=udmDOaHN9no' 'https://www.youtube.com/watch?v=7CTH59GCNko'

For windows, the program might be blocked by your antivirus, be aware of this possibility.

Audio feedback

By default, the video_creator_main.py script uses the tycat command, available on Terminology terminal emulator to play the audio feedback while building the video.

If you want to use any other program, give it to argument with -c. Replace the filename in the command by {}.

Example:
python video_creator_main.py -c "vlc --play-and-exit --no-repeat --no-loop {}" config.json 'https://www.youtube.com/watch?v=udmDOaHN9no'

For Windows release

You will face a problem if the command you want to use contains a space in its path.

Here is a simple trick to work around the problem. Let's say we want to use vlc as audio feedback command.
Create feedback.bat with the following content:

"C:\Program Files\VideoLAN\VLC\vlc.exe" --play-and-exit --no-repeat --no-loop %*

(Path to vlc might change depending of your setup)

Then, call the program with the followin -c parameter:
.\SentenceMixerCLI.exe config.json 'https://www.youtube.com/watch?v=VTaoMHjnRbM' -c ".\feeback.bat {}"

Interactive process

The audio sequence creation process works as follows:

1. Enter the chunk of the sentence you want to hear
2. You will hear an audio feedback from the chosen combo. There you can press:
   • ENTER if you are not satisfied with current combo for this chunk and want another phonemes
   • e to change the written chunk
   • s to store the current combo in a buffer, so you can browse other combos without fearing to lose this one
   • l + number to load a previously stored combo
   • y to accept the current combo and go to next chunk
3. Loop to 1. until you enter an empty chunk
4. The full video is formed from the association of all the selected combo

Tips

• Punctuation symbols count as blank phonemes. It can be very useful to add pause in your sentences.
• Do not hesitate to put a lot of Youtube URLs in argument. The more the video material, the more choice you will have to form the sentences you want.
• An already downloaded video will not be downloaded again. So, do not hesitate then to download long videos to boost accuracy.
• Advance little by little. There is no big difference in accuracy between long and short chunks, so, it is generally preferable to write short chunks (+/- 1 word by chunk) to have more control of how each word/syllabus sound. For example, it is probably better to split sentence "Salut les gens" in three chunks "Salut", "les", "gens" than in one big chunk "Salut les gens".
• For some long words, it can sometimes be interesting to split it in little chunks. For example, you can split the word Compagnie in three chunks "Con", "Pas", "Nie".
• If a word you want to hear can be directly heard in one of the source video (same word or homophone), don't split it in chunks, it will directly been taken from the source video. For example, if the word "Camp" is present in source subtitles, words "Camp", "Quand", "Quant" will automatically associate to the original "Camp".
• Related to previous point, try to use a maximum of words you can find in the original video. The result will be way more accurate and understandable.

Restrictions and workarounds

Subtitles free videos

Only subtitled youtube videos are usable. Automatically generated youtube subtitles work as well.
If you really want to use a non subtitled video, you will have to write the subtitles yourself.

Phoneme not found

If your source video corpus is very short or contains very few words, it might be possible that some phonemes are not prounounced at all through the whole video material. In that case, you will not be able to use these particular phonemes.
The problem can be easily solved by adding video material to the corpus, or not using these missing phonemes.

Unrecognized word

You can only enter a word that already exists in the dictionnary. If you want to hear an external word, you have two solutions:

1. Add the word in the dictionnary
2. Split the word in sub-dictionnary-registered words

For example, the word "Bolos" can be either:

1. Added in the dictionnary by adding the line "BOLOS b o l o s"
2. Decomposed in three valid words "Beau" "l'eau" "se"

Ambiguity intolerance

Unfortunately, the system doesn't tolerates ambiguous words for the moment.
For exemple, word "Est" can be pronounced:

1. With phoneme "e" (to be)
2. With phonemes "E s t" (east)

Thus, it is for the moment forbidden to write down word "Est". Waiting for a disambiguation feature, just change the word for an homophone.

For word "Est", if you want to hear phoneme "e", you can use word "Et" ; if you want to hear phonemes "e s t", you can use the sequence of words "S" "te" (not perfect coverage of phonemes in this case however).

Unknown encoding idna

This error is particulary serious and randomly shows when using an outdated version of youtube-dl.
If you have installed the application from source, you can simply update youtube-dl via pip.
There is no solution for the moment for release versions. The only thing you can do is to install the program from source.
We are looking for a solution concerning this problem .

Unsupported languages

Please refer to this section of the Sentence Mixing library documentation.

Building project

You can make your own release of the program in order to use it freely without dependancies.
Everything concerning this feature is explained in folder builder.