T2TT in 96 languages?

#14
by M2LabOrg - opened

I have been testing seamless-m4t-v2-large following the example given on Hugging Face. I am trying to translate "Hello, my dog is cute" in all of the 96 languages. However, only 35 languages seem to work. Why is that?

Here's my code:

from transformers import AutoProcessor, SeamlessM4Tv2Model
import torch
import soundfile
import os
import json

# Load language codes and names from JSON file
with open('languages.json', 'r') as file:
    data = json.load(file)
    language_codes = list(data['languages'].keys())
    language_names = data['languages']

# Load the Seamless model and processor
processor = AutoProcessor.from_pretrained("facebook/seamless-m4t-v2-large")
model = SeamlessM4Tv2Model.from_pretrained("facebook/seamless-m4t-v2-large")

# Set the device to CUDA if available, else use CPU
device = "cuda:0" if torch.cuda.is_available() else "cpu"

# List of target language codes
language_codes = [
    'afr', 'amh', 'arb', 'ary', 'arz', 'asm', 'ast', 'azj', 'bel', 'ben',
    'bos', 'bul', 'cat', 'ceb', 'ces', 'ckb', 'cmn', 'cmn_Hant', 'cym', 'dan',
    'deu', 'ell', 'eng', 'est', 'eus', 'fin', 'fra', 'fuv', 'gaz', 'gle',
    'glg', 'guj', 'heb', 'hin', 'hrv', 'hun', 'hye', 'ibo', 'ind', 'isl', 'ita',
    'jav', 'jpn', 'kam', 'kan', 'kat', 'kaz', 'kea', 'khk', 'khm', 'kir', 'kor',
    'lao', 'lit', 'ltz', 'lug', 'luo', 'lvs', 'mai', 'mal', 'mar', 'mkd', 'mlt',
    'mni', 'mya', 'nld', 'nno', 'nob', 'npi', 'nya', 'oci', 'ory', 'pan', 'pbt',
    'pes', 'pol', 'por', 'ron', 'rus', 'slk', 'slv', 'sna', 'snd', 'som', 'spa',
    'srp', 'swe', 'swh', 'tam', 'tel', 'tgk', 'tgl', 'tha', 'tur', 'ukr', 'urd',
    'uzn', 'vie', 'xho', 'yor', 'yue', 'zlm', 'zsm', 'zul'
]

text_to_translate = "Hello, my dog is cute"

# Translate and print results
for tgt_lang in language_codes:
    try:
        text_inputs = processor(text=text_to_translate, src_lang="eng", return_tensors="pt").to(device)
        decoder_input_ids = model.generate(**text_inputs, tgt_lang=tgt_lang, generate_speech=False)[0].tolist()
        translated_text = processor.decode(decoder_input_ids[0], skip_special_tokens=True)

        print(f"=> {tgt_lang}: {translated_text}")

        audio_array_from_text = model.generate(**text_inputs, tgt_lang=tgt_lang)[0].cpu().numpy().squeeze()
        sample_rate = model.config.sampling_rate
        soundfile.write(f"audio_output/{tgt_lang}_output_audio.wav", audio_array_from_text, sample_rate)

    except ValueError as ve:
        if "is not supported by this model" in str(ve):
            continue
        else:
            print(f"Error translating to {tgt_lang}: {ve}")
            continue
    except Exception as e:
        print(f"Error translating to {tgt_lang}: {e}")
        continue

And here's the output:

=> arb: مرحباً، كلبي حلو
=> ben: হ্যালো, আমার কুকুরটা সুন্দর
=> cat: Hola, el meu gos és bonic.
=> ces: Ahoj, můj pes je roztomilý.
=> cmn: 你好,我的狗很可爱
=> cym: Helo, mae fy nghŵn yn ddoniol
=> dan: Hej, min hund er sød.
=> deu: Hallo, mein Hund ist süß.
=> eng: Hello, my dog is cute
=> est: Tere, mu koer on armas.
=> fin: Hei, koirani on söpö.
=> fra: Salut, mon chien est mignon
=> hin: हैलो, मेरे कुत्ते प्यारा है
=> ind: Halo, anjingku lucu
=> ita: Ciao, il mio cane è carino.
=> jpn: こんにちは私の犬は可愛いです
=> kor: 안하세요, 제 개는 귀습니다.
=> mlt: Hello, il-kelb tiegħi huwa ħelu
=> nld: Hallo, mijn hond is schattig.
=> pes: سلام، سگ من دوست داشتنی است.
=> pol: Cześć, mój pies jest uroczy.
=> por: Olá, o meu cão é bonito.
=> ron: Bună ziua, câinele meu este drăguţ.
=> rus: Привет, моя собака милая.
=> slk: Ahoj, môj pes je roztomilý.
=> spa: Hola, mi perro es lindo.
=> swe: Hallå, min hund är söt.
=> swh: Hello, mbwa wangu ni cute
=> tel: హలో, నా కుక్క అందమైన ఉంది
=> tgl: Kumusta, ang aking aso ay kaakit-akit
=> tha: สวัสดีครับ สุนัขผมน่ารักมาก
=> tur: Merhaba, köpeğim çok tatlı.
=> ukr: Привіт, мій собака мила.
=> urd: ہیلو، میرا کتا پیارا ہے
=> uzn: Assalomu alaykum, mening itim juda yoqimli.
=> vie: Xin chào, con chó của tôi rất dễ thương.

M2LabOrg changed discussion title from T2TT in 96 languages to T2TT in 96 languages?
AI at Meta org

There are actually 98 text languages supported (we didn't count languages with 2 scripts like cmn/cmn_Hant or the two Norwegian varieties nno/nob).
This is the script I'm using to generate T2TT outputs in all 98 supported languages:

from transformers import AutoProcessor, SeamlessM4Tv2Model, SeamlessM4Tv2ForTextToText
import torch

# Set the device to CUDA if available, else use CPU
device = "cuda:0" if torch.cuda.is_available() else "cpu"

# Load the Seamless model and processor
processor = AutoProcessor.from_pretrained("facebook/seamless-m4t-v2-large")
model = SeamlessM4Tv2ForTextToText.from_pretrained("facebook/seamless-m4t-v2-large").to(device)

# List of target language codes
language_codes = [
    'afr', 'amh', 'arb', 'ary', 'arz', 'asm', 'azj', 'bel',
    'ben', 'bos', 'bul', 'cat', 'ceb', 'ces', 'ckb', 'cmn',
    'cmn_Hant', 'cym', 'dan', 'deu', 'ell', 'eng', 'est',
    'eus', 'fin', 'fra', 'fuv', 'gaz', 'gle', 'glg', 'guj',
    'heb', 'hin', 'hrv', 'hun', 'hye', 'ibo', 'ind', 'isl',
    'ita', 'jav', 'jpn', 'kan', 'kat', 'kaz', 'khk', 'khm',
    'kir', 'kor', 'lao', 'lit', 'lug', 'luo', 'lvs', 'mai',
    'mal', 'mar', 'mkd', 'mlt', 'mni', 'mya', 'nld', 'nno',
    'nob', 'npi', 'nya', 'ory', 'pan', 'pbt', 'pes', 'pol',
    'por', 'ron', 'rus', 'sat', 'slk', 'slv', 'sna', 'snd',
    'som', 'spa', 'srp', 'swe', 'swh', 'tam', 'tel', 'tgk',
    'tgl', 'tha', 'tur', 'ukr', 'urd', 'uzn', 'vie', 'yor',
    'yue', 'zlm', 'zul'
]

text_to_translate = "Hello, my dog is cute"

# Translate and print results
for tgt_lang in language_codes:
    text_inputs = processor(text=text_to_translate, src_lang="eng", return_tensors="pt").to(device)
    decoder_input_ids = model.generate(**text_inputs, tgt_lang=tgt_lang)[0].tolist()
    translated_text = processor.decode(decoder_input_ids, skip_special_tokens=True)
    print(f"=> {tgt_lang}: {translated_text}")

These are the outputs I get:

=> afr: Hallo, my hond is oulik
=> amh: ጤና ይስጥልኝ፣ ውሻዬ በጣም ቆንጆ ነው።
=> arb: مرحباً، كلبي حلو
=> ary: أهلا، الكلب ديالي حلو.
=> arz: أهلا، كلبي حلو.
=> asm: হেল্লো, মোৰ কুকুৰাটো মৰমলগা
=> azj: Salam, mənim köpəyim çox şirindir.
=> bel: Прывітанне, мой сабака мілы
=> ben: হ্যালো, আমার কুকুরটা সুন্দর
=> bos: Zdravo, moj pas je slatki.
=> bul: Здравей, кучето ми е сладко.
=> cat: Hola, el meu gos és bonic.
=> ceb: Kumusta, ang akong iro matahom
=> ces: Ahoj, můj pes je roztomilý.
=> ckb: سڵاو، سەگەکەم زۆر خۆشە
=> cmn: 你好,我的狗很可爱
=> cmn_Hant: 大家好,我的狗很可愛
=> cym: Helo, mae fy nghŵn yn ddoniol
=> dan: Hej, min hund er sød.
=> deu: Hallo, mein Hund ist süß.
=> ell: Γεια σου, ο σκύλος μου είναι χαριτωμένος.
=> eng: Hello, my dog is cute
=> est: Tere, mu koer on armas.
=> eus: Kaixo, nire txakurra oso polita da.
=> fin: Hei, koirani on söpö.
=> fra: Salut, mon chien est mignon
=> fuv: Hello, my dog is cute. (Hello, my dog is cute.)
=> gaz: Salaam, sareen koo baay'ee bareedaa dha
=> gle: Dia duit, tá mo mhadra suimiúil
=> glg: Ola, o meu can é bonito.
=> guj: હેલો, મારો કૂતરો સુંદર છે
=> heb: שלום, הכלב שלי חמוד.
=> hin: हैलो, मेरे कुत्ते प्यारा है
=> hrv: Zdravo, moj pas je slatki.
=> hun: Helló, a kutyám aranyos.
=> hye: Ողջույն, իմ շունը քաղցր է:
=> ibo: Ndewo, nkịta m mara mma
=> ind: Halo, anjingku lucu
=> isl: Halló, hundurinn minn er sætur.
=> ita: Ciao, il mio cane è carino.
=> jav: Halo, segawonku lucu banget
=> jpn: こんにちは私の犬は可愛いです
=> kan: ಹಲೋ, ನನ್ನ ನಾಯಿ ಮುದ್ದಾಗಿದೆ
=> kat: ჰელო, ჩემი ძაღლი საყვარელია.
=> kaz: Сәлем, менің итім сүйкімді.
=> khk: Сайн байна уу, миний нохой гоё байна.
=> khm: សួស្តី ឆ្កែ របស់ខ្ញុំ ស្រស់ស្អាត ណាស់
=> kir: Саламатсызбы, менин итим сүйкүмдүү.
=> kor: 안하세요, 제 개는 귀습니다.
=> lao: ສະບາຍດີ, ຫມາ ຂອງຂ້ອຍຫນ້າຮັກ
=> lit: Sveiki, mano šuo yra mielas.
=> lug: Hello, embwa yange nnungi nnyo
=> luo: Hello, guok mara ber ahinya
=> lvs: Sveiki, mans suns ir jauks.
=> mai: हेलो, हमर कुत्ता बहुत प्यारा अछि
=> mal: ഹലോ, എന്റെ നായ സുന്ദരമാണ്.
=> mar: हॅलो, माझा कुत्रा गोड आहे
=> mkd: Здраво, моето куче е слатко.
=> mlt: Hello, il-kelb tiegħi huwa ħelu
=> mni: হেলো, ঐগী শগোল অসি য়াম্না নুংশিজৈ
=> mya: ဟဲလို၊ ငါ့ခွေးက ချစ်စရာကောင်းတယ်
=> nld: Hallo, mijn hond is schattig.
=> nno: Hallå, hunden min er søt.
=> nob: Hei, hunden min er søt.
=> npi: नमस्कार, मेरो कुकुर प्यारा छ
=> nya: Moni, galu wanga ndi wokongola
=> ory: ହେଲୋ, ମୋର କୁକୁରଟି ବହୁତ ସୁନ୍ଦରୀ
=> pan: ਹੈਲੋ, ਮੇਰਾ ਕੁੱਤਾ ਪਿਆਰਾ ਹੈ
=> pbt: سلام، زما سپي ډېر ښکلی دی
=> pes: سلام، سگ من دوست داشتنی است.
=> pol: Cześć, mój pies jest uroczy.
=> por: Olá, o meu cão é bonito.
=> ron: Bună ziua, câinele meu este drăguţ.
=> rus: Привет, моя собака милая.
=> sat: হেলো, ঐগী শগোল য়াম্না মধুর
=> slk: Ahoj, môj pes je roztomilý.
=> slv: Pozdravljeni, moj pes je srčkan.
=> sna: Mhoroi, imbwa yangu inotapira
=> snd: هيلو، منهنجو ڪتو پيارو آهي
=> som: Salaan, eeyga waa mid qurux badan
=> spa: Hola, mi perro es lindo.
=> srp: Здраво, мој пас је слатки.
=> swe: Hallå, min hund är söt.
=> swh: Hello, mbwa wangu ni cute
=> tam: ஹலோ, என் நாய் அழகாக இருக்கிறது
=> tel: హలో, నా కుక్క అందమైన ఉంది
=> tgk: Салом, саги ман хеле зебо аст.
=> tgl: Kumusta, ang aking aso ay kaakit-akit
=> tha: สวัสดีครับ สุนัขผมน่ารักมาก
=> tur: Merhaba, köpeğim çok tatlı.
=> ukr: Привіт, мій собака мила.
=> urd: ہیلو، میرا کتا پیارا ہے
=> uzn: Assalomu alaykum, mening itim juda yoqimli.
=> vie: Xin chào, con chó của tôi rất dễ thương.
=> yor: Ẹ kú o, ajá mi dùn ún wò.
=> yue: 大家好,我個狗好可愛
=> zlm: Hello, anjing saya sangat comel
=> zul: Sawubona, inja yami imnandi

@elbayadm Just wanted to drop a quick thank you – everything's running smooth now, and I really appreciate your help with SeamlessM4T! You rock!

M2LabOrg changed discussion status to closed

Sign up or log in to comment