gruut is too slow

[yiershanxll]

when I use gruut to convert arabic text to phomes. It cost me 6s / 10 sentence, sentence is same like "'الرئيس مريدي والبيادي وأقصد تماما ليبيا'", my question is how to improve the speed

[neurlang]

hello, I started a competing project. The upstream was not responsive about adding my language. I plan to add Arabic and it will be faster.

[neurlang]

Hello, I can phonemize sentence like that in 3ms. The quality isn't the best.

{
	"Words": [
		{
			"Linguistic": "الرئيس",
			"Phonetic": "ːrlirarjasa"
		},
		{
			"Linguistic": "مريدي",
			"Phonetic": "maraːdai"
		},
		{
			"Linguistic": "والبيادي",
			"Phonetic": "ːtʔalabiːbadaja"
		},
		{
			"Linguistic": "وأقصد",
			"Phonetic": "waʔaqasˤdi"
		},
		{
			"Linguistic": "تماما",
			"Phonetic": "tamuʔamːb"
		},
		{
			"Linguistic": "ليبيا",
			"Phonetic": "liːbiːbʔa"
		}
	]
}

[neurlang]

Tried today, time was 683 ms

{
   "Words":[
      {
         "Linguistic":"الرئيس",
         "Phonetic":"ːlararjasa"
      },
      {
         "Linguistic":"مريدي",
         "Phonetic":"marajadaja"
      },
      {
         "Linguistic":"والبيادي",
         "Phonetic":"waːlabijaːdaja"
      },
      {
         "Linguistic":"وأقصد",
         "Phonetic":"waʔaqasˤda"
      },
      {
         "Linguistic":"تماما",
         "Phonetic":"tamaːmaː"
      },
      {
         "Linguistic":"ليبيا",
         "Phonetic":"lajabijaː"
      }
   ]
}

Rating by chat gpt:

Overall Quality Rating:

The phonetic transcription quality is inconsistent, with significant errors in several entries. The phonetic transcriptions need considerable improvement.

Overall Rating: Poor to Fair.

[synesthesiam]

Arabic uses mishkal to preprocess the text to add diacritics, which is slow.

It will be faster but less accurate if mishkal isn't installed.

[neurlang]

Today's retrain had this result:

ælrːjs marjːdj ˈhælbjɑːdj ˈhʔaqsˈːd ːtmaɑmaɑ liːɣajbjɑ

Chat confirms its the most accurate so far.

[martinarisk]

Goruut 0.4.0 says: rajsˈim mrjdiː waːlbiɑdj uːaqsd tamaːmaː liːbijɑ Overall Assessment:

While none of the options perfectly capture the standard IPA for the phrase, Option A is the least problematic overall. It handles “والبيادي,” “تماما,” and “ليبيا” in a relatively acceptable way, even though it falls short on “الرئيس” and “وأقصد.” In contrast, Options B, C, and D introduce additional sounds or extra syllables that stray farther from the expected pronunciations.

Conclusion: Option A is the best of the four, even if it isn’t perfect.

[martinarisk]

Goruut 0.6.1 says: rajsˈim mriːdi waːlabaːaːdˈi waqsˤd tamaːmaː liːbijɑ