Why does Tamil voice recognition produce more errors than other Indian languages?

Tamil's diglossia is the primary cause. ASR models are trained on written centamil, but speakers naturally use spoken koduntamil — which has completely different verb forms, pronouns, and phonology. 'போகிறேன்' (written) vs 'போறேன்' (spoken) for 'I am going.' Dictating in centamil (formal written Tamil) reduces error rates by 40–60% compared to natural spoken Tamil.

Does Tamil voice to text work for Sri Lankan Tamil (Jaffna)?

Poorly in natural Jaffna Tamil. Sri Lankan Tamil preserves phonological distinctions and archaic vocabulary that diverge from Indian centamil. There is no dedicated Sri Lankan Tamil locale. Jaffna speakers get best results by approximating Indian centamil verb forms and vocabulary.

What is the ழ (zha) sound and why does it matter for Tamil ASR?

ழ represents a retroflex approximant /ɻ/ unique to Tamil and Malayalam — found in no other major world language. The word Tamil itself ends with this sound. If produced as /l/ or /r/ instead, the model outputs ல or ர. Consciously producing the retroflex position improves recognition of the ~400+ Tamil words containing ழ.

What makes Tamil script unique compared to other Indian scripts for ASR?

Tamil script has no capital letters (proper nouns get no visual marker), no loan-script characters for foreign sounds (English words are fully adapted into Tamil phonology), and encodes a six-way place-of-articulation distinction in stop consonants. Stop consonants are also context-dependent allophones — the same character is voiced or voiceless depending on position.

Tamil Voice to Text — தமிழ் குரல் உரை மாற்றி

Q: தமிழ் குரல் உரை மாற்றம் mobile-ல் வேலை செய்யுமா?

ஆம். Android Chrome மற்றும் iPhone Safari-ல் வேலை செய்கிறது. Android Chrome தமிழுக்கு சிறந்த முடிவுகளை தருகிறது. எந்த app-ஐயும் நிறுவ தேவையில்லை.

Q: Can I use Tamil voice to text in Tanglish (Tamil-English mix)?

Yes. English words are transliterated into Tamil script — 'meeting' becomes 'மீட்டிங்.' Tamil case markers attach correctly to English words ('meeting-ல,' 'file-ஐ'). For formal content with English in Roman script, type English terms manually during dictation pauses.

Tamil is one of the world's oldest living languages — with a continuous literary tradition stretching back over 2,000 years — and one of the most linguistically complex for speech recognition. It has a diglossia more extreme than Arabic's: the formal written register (centamil, செந்தமிழ்) and the spoken colloquial language (koduntamil, கொடுந்தமிழ்) differ so substantially that a Tamil speaker dictating naturally may produce output the model barely recognises. Tamil script has no capital letters and encodes a six-way place-of-articulation distinction in its stop consonants that no other major language maintains. And Tamil spoken in Chennai, Jaffna, Kuala Lumpur, and Singapore are distinct enough to require different recognition strategies. This page covers all of it.

தமிழ் குரல் அங்கீகாரத்தில் மிகப் பெரிய சவால் செந்தமிழ் மற்றும் கொடுந்தமிழ் இடையேயான வேறுபாடு. இந்த பக்கம் குரல் அங்கீகாரம் எவ்வாறு செயல்படுகிறது என்பதையும் சிறந்த முடிவுகளை எவ்வாறு பெறுவது என்பதையும் விளக்குகிறது.

Tamil Diglossia: The Widest Formal/Spoken Gap of Any Living Language

Tamil diglossia — the coexistence of centamil (literary/formal Tamil) and koduntamil (spoken/colloquial Tamil) — is considered by linguists to be one of the most extreme examples of diglossia in any living language. The two varieties differ not just in vocabulary and style but in fundamental grammar: verb paradigms, case suffixes, pronouns, and negation patterns are entirely different between the two registers.

Meaning	செந்தமிழ் Centamil (written/formal)	கொடுந்தமிழ் Koduntamil (spoken)
I am going	நான் போகிறேன்	நான் போறேன்
He said	அவன் சொன்னான்	அவன் சொன்னான் (same — past less affected)
What is this?	இது என்ன?	இது என்னா?
I don't know	எனக்குத் தெரியாது	எனக்கு தெரியல
Come here	இங்கே வாருங்கள்	இங்க வா
They are eating	அவர்கள் சாப்பிடுகிறார்கள்	அவங்க சாப்பிடுறாங்க
He/She (respectful)	அவர்	அவரு / அவுங்க
I (first person)	நான்	நான் (same — pronoun stable)

Speech recognition models are trained predominantly on written Tamil text — which is centamil. When you speak naturally in koduntamil (as virtually all Tamil speakers do in daily life), the model hears phoneme sequences that don't match its training data. "போறேன்" (spoken "going") has a different vowel and consonant pattern from "போகிறேன்" (written "going"). The model may output the written form, a corrupted form, or fail to recognise the word entirely.

Most important practical tip

For best Tamil voice recognition accuracy, dictate in centamil — formal written Tamil — even though it's not how you naturally speak. Say "போகிறேன்" not "போறேன்." Say "சாப்பிடுகிறார்கள்" not "சாப்பிடுறாங்க." This single change can reduce error rates by 40–60% compared to natural spoken Tamil.

Tamil Script — தமிழ் எழுத்து: Unique Characteristics

Tamil script is an abugida — each character represents a consonant with an inherent vowel, modified by diacritic vowel signs. But it has several features that make it distinct from Devanagari, Bangla, or any other Indian script:

🔡 No Capital Letters

Tamil script has no uppercase/lowercase distinction — every character has a single form. This means proper nouns (names, cities, countries) are written in the same script as common words, with no visual marker. "சென்னை" (Chennai), "குமார்" (Kumar), and "தமிழ்" (Tamil) look the same structurally as any other noun. The model must use context to identify proper nouns — it cannot rely on capitalisation as a signal, unlike Latin-script languages.

🚫 No Loan-Script Characters

Unlike Hindi (which uses nuktā-modified Devanagari characters to represent Persian/Arabic sounds) or Bengali (which has similar mechanisms), Tamil traditionally does not have special characters for foreign sounds. English loanwords are adapted to the existing Tamil phonological inventory — "computer" becomes "கம்ப்யூட்டர்" using only native Tamil characters. This means English words in Tamil speech produce Tamil-script output, not Roman-script output.

🔢 Tamil Numerals

Tamil has its own numeral script: ௧ ௨ ௩ ௪ ௫ ௬ ௭ ௮ ௯ ௰ (1–10). In modern digital Tamil writing, Western Arabic numerals (1, 2, 3) are universally used — Tamil numerals appear only in traditional contexts, classical literature, and certain religious texts. Speech recognition outputs Western Arabic numerals by default. If you need traditional Tamil numerals for heritage or classical content, replace them after dictation.

⊕ Pulli — The Vowel Suppressor

The pulli (்) is a small dot placed above a Tamil consonant character to suppress its inherent vowel, indicating a pure consonant. "கம்ப்யூட்டர்" (computer) has multiple pullis marking consonant clusters. The model outputs pulli correctly for common loanwords. For less common foreign-origin words adapted into Tamil, pulli placement may be inconsistent — check technically complex loanwords manually.

🔀 Grantha Characters

Modern Tamil script includes a set of "Grantha" characters — ஜ (ja), ஷ (sha), ஸ (sa), ஹ (ha), க்ஷ (ksha) — borrowed from Sanskrit to represent sounds not native to classical Tamil. These are used in Sanskrit-origin words and some English loanwords. The model outputs Grantha characters correctly for words where their use is established (ஜனவரி for January, ஷார்ட் for short). In puristic Tamil writing that avoids Grantha characters, these are replaced with native Tamil equivalents — but models output the more common Grantha forms.

📏 Āytam — The Rare Third Category

Tamil has a unique character called the āytam (ஃ) — classified as neither vowel nor consonant. It appears in a small number of words and represents a fricative sound. "அஃது" (that thing — archaic), "கஃடு" (in some dialects). In modern Tamil, āytam is rare in everyday writing and speech recognition models handle it only in the few words where it is conventional — it will not appear spuriously in normal dictation output.

Tamil's Six-Place Stop System — ஆறு இட நிறுத்தங்கள்

Tamil has one of the most elaborate stop consonant systems of any language. Where English distinguishes stops at three places of articulation (bilabial p/b, alveolar t/d, velar k/g), Tamil distinguishes stops at six places:

Place of Articulation	Tamil Character	Romanisation	English approximation
Bilabial	ப	p	p in "spin" (unaspirated)
Alveolar	த	t	dental t — tip of tongue on teeth
Retroflex	ட	ṭ	tongue curled back — no English equivalent
Palatal	ச	c/ch	between ch and s
Velar	க	k	k in "skin" (unaspirated)
Alveolar lateral	ற	ṟ/r	a tapped/trilled r unique to Tamil

Crucially, Tamil does not distinguish voiced from voiceless in its native stops — the same character ப (pa) can be pronounced /p/, /b/, or /β/ depending on position in a word. This context-dependent allophony means the model must determine the correct character from position and context, not from the acoustic signal alone. The retroflex/dental distinction (ட vs த) is particularly challenging — non-native speakers and speakers from certain regional varieties may not maintain this distinction clearly in fast speech.

Tamil Dialect Accuracy by Region — மாவட்ட வழக்கு

Tamil is spoken across a vast geographic spread — Tamil Nadu and Puducherry in India, Sri Lanka (particularly the Northern Province), Malaysia, Singapore, and diaspora communities in the UK, Canada, and South Africa. The dialectal variation is significant, and ASR models are trained primarily on Tamil Nadu standard Tamil:

🏙️

Chennai Standard — Best Results

The educated urban Tamil of Chennai — used in Tamil media, cinema, and formal settings — is the reference model for Tamil ASR. This is the variety of Kollywood (Tamil film industry), major TV channels, and news broadcasting. Dictating in Chennai-standard Tamil with centamil verb forms gives the best accuracy. Expect word error rates of 10–15% in quiet conditions with clear speech.

🏰

Madurai / Tirunelveli — Moderate

Southern Tamil (Madurai, Tirunelveli, Thoothukudi) has distinctive features: the "a" vowel in certain positions becomes more open, retroflex consonants are more strongly articulated, and the colloquial register differs substantially from Chennai Tamil. "வருவான்" (he will come) may become "வருவாண்டா" in Madurai speech. Models handle educated southern Tamil acceptably; strong local dialect features cause higher errors.

🏭

Coimbatore / Salem — Moderate

Western Tamil Nadu (Coimbatore, Salem, Erode) has its own distinctive intonation pattern and some vocabulary differences. The Kongu Tamil variety — spoken in this region — has features influenced by Kannada and Telugu proximity. Error rates are moderate for standard speech; heavy Kongu Tamil features increase errors.

🇱🇰

Sri Lankan Tamil (Jaffna) — Challenging

Jaffna Tamil — the prestige variety of Sri Lankan Tamil — has maintained phonological distinctions that have collapsed in most Indian Tamil varieties, including a distinction between short and long consonants that Indian Tamil has lost. It preserves archaic vocabulary, has distinct intonation patterns, and its formal register (closer to classical Tamil) differs from Indian Tamil's centamil. Standard Indian Tamil ASR models perform poorly on Jaffna Tamil. Sri Lankan Tamil speakers should dictate as close to centamil as possible.

🇲🇾🇸🇬

Malaysian & Singapore Tamil — Challenging

Tamil spoken in Malaysia and Singapore has been evolving in contact with Malay, English, Chinese languages, and each other for generations. Malaysian Tamil has Malay loanwords; Singapore Tamil mixes Tamil, English, Malay, and Hokkien. Both are heavily code-switching varieties. Standard Indian Tamil ASR handles the Tamil portions of formal Malaysian/Singapore Tamil speech moderately; the heavy English/Malay mixing causes significant errors. Dictate the Tamil portions in centamil and type the non-Tamil portions.

🇬🇧🇨🇦

Diaspora Tamil (UK, Canada) — Variable

Tamil diaspora communities in the UK (particularly London) and Canada (particularly Toronto's Scarborough neighbourhood) include both Sri Lankan Tamil and Indian Tamil speakers. Sri Lankan diaspora Tamil has maintained Jaffna features; Indian diaspora Tamil varies. English code-switching is very heavy. For dictation purposes, the Tamil portions work best in centamil with English terms typed separately.

எவ்வாறு பயன்படுத்துவது — How to Start

மொழி மெனுவிலிருந்து "Tamil (தமிழ்)" அல்லது "ta-IN" தேர்ந்தெடுக்கவும்

Select Tamil (ta-IN) from the language menu. Chrome on desktop has the best Tamil ASR support.

"Start 🎤" பொத்தானை அழுத்தி மைக்ரோஃபோன் அனுமதி வழங்கவும்

Click Start and allow microphone access. Quiet environment is especially important for Tamil — retroflex/dental confusion increases with background noise.

முடிந்தவரை செந்தமிழில் பேசுங்கள் — "போகிறேன்" not "போறேன்", "சாப்பிடுகிறேன்" not "சாப்பிடுறேன்"

This is the single most important tip: dictate in centamil (formal Tamil) even though it's not your natural speech. Error rates drop 40–60% compared to spoken koduntamil.

உரையை நகலெடுக்கவும் அல்லது TXT கோப்பாக பதிவிறக்கவும். தமிழ் எழுத்துக்கள் WhatsApp மற்றும் Word இல் சரியாகக் காட்டப்படும்

Copy text or download as TXT. Tamil script renders correctly in WhatsApp, Word, and all modern apps. Check pulli placement in complex loanwords manually.

Tanglish: Tamil-English Code-Switching

Urban Tamil — particularly in Chennai, Bangalore's Tamil community, and the global Tamil diaspora — mixes Tamil and English so thoroughly that it has earned its own name: Tanglish. A typical Chennai professional's speech might be:

"அந்த meeting-ல என்ன discuss பண்ணாங்க?"

"Deadline நாளைக்கு இருக்கு, file-ஐ share பண்ணுங்க."

"Project எப்படி progress ஆகுது?"

Purple = English words; note Tamil case markers attached: file-ஐ (accusative), meeting-ல (locative)

Note the characteristic Tamil case suffixes attached directly to English words: "file-ஐ" (the file, accusative case), "meeting-ல" (in the meeting, locative case). Tamil inflects English words with Tamil grammatical suffixes — and the ASR model in ta-IN mode handles this correctly for common English words.

English words will be transliterated into Tamil script — "meeting" becomes "மீட்டிங்," "deadline" becomes "டெட்லைன்." For formal content where you want English in Roman script, pause dictation and type English terms manually. For WhatsApp and informal notes, Tamil-script transliteration of common English words is serviceable.

Tamil Phonology: Key ASR Challenges

🔄 Context-Dependent Allophones

Tamil stops have no voicing distinction in phonemic inventory — the same character is pronounced differently depending on position. ப (pa) is /p/ word-initially, /b/ or /β/ intervocalically, and /p/ or /b/ in clusters. The model must output the correct character (ப) regardless of which allophone it hears. This is generally handled correctly — but fast speech where allophonic variation is exaggerated increases errors.

🌀 The ழ (zha) — Tamil's Unique Sound

The letter ழ (romanised as "zh" or "ḻ") represents a sound found only in Tamil and Malayalam — a retroflex approximant /ɻ/ that has no equivalent in any other major world language. The word "தமிழ்" (Tamil) itself ends with this sound. Non-native Tamil speakers (and some native speakers in northern dialects) may substitute /l/ or /r/ for ழ, causing the model to output ல or ர instead. Consciously producing the retroflex approximant improves recognition of words containing ழ.

🔤 ன vs ண vs ங — Three Nasals

Tamil distinguishes three nasal consonants at different places of articulation: ன (alveolar /n/), ண (retroflex /ɳ/), and ங (velar /ŋ/). In spoken Tamil these distinctions are often neutralised — particularly the retroflex/alveolar distinction. The model must output the correct nasal character based on word-level knowledge. Errors occur in words where the nasal type is not acoustically distinct — check ன/ண in formal writing.

📏 Vowel Length — Short vs Long

Tamil systematically distinguishes short and long vowels — five short vowels (அ, இ, உ, எ, ஒ) and five long counterparts (ஆ, ஈ, ஊ, ஏ, ஓ), plus two diphthongs (ஐ, ஔ). Vowel length is phonemic — "பல" (pala, many) vs "பால்" (pāl, milk). In fast speech, the length distinction is reduced. The model uses lexical context to output the correct vowel — it usually succeeds for common words but may confuse length-distinguished pairs in less frequent vocabulary.

சிறந்த துல்லியத்திற்கான குறிப்புகள் — Tips for Best Accuracy

✅ துல்லியத்தை மேம்படுத்துவது

• செந்தமிழில் பேசுங்கள் — இது மிகவும் முக்கியமான குறிப்பு
• "போகிறேன்" not "போறேன்" — formal verb forms use
• ழ-வை தெளிவாக உச்சரியுங்கள் — retroflex approximant
• சரியான ட/த வேறுபாட்டை பராமரிக்கவும்
• முழு வாக்கியத்தை முடித்த பின்னர் இடைவேளை எடுங்கள்
• Chrome browser-ல் சிறந்த தமிழ் ASR கிடைக்கும்
• English வார்த்தைகளை இடைவேளையில் type செய்யுங்கள்

⚠️ பொதுவான பிழைகளும் தீர்வுகளும்

• கொடுந்தமிழ் verb forms — formal forms-ல் திரும்பி type செய்யவும்
• English in Tamil script — pause செய்து Roman-ல் type
• ன/ண confusion — formal writing-ல் manual check
• ழ vs ல/ர substitution — retroflex-ஐ தெளிவாக உச்சரிக்கவும்
• Jaffna/Malaysian Tamil — Chennai standard-ல் பேசவும்
• Vowel length pairs (பல/பால்) — uncommon words check

யார் பயன்படுத்துகிறார்கள் — Who Uses Tamil Voice to Text

📱

WhatsApp & Messaging

Tamil WhatsApp usage is enormous across Tamil Nadu, Sri Lanka, and the diaspora. Typing Tamil on a touchscreen — navigating Tamil keyboard layouts, handling vowel signs, selecting the right conjunct — is significantly slower than typing English. Voice dictation for Tamil WhatsApp messages is 4–5× faster for most users.

🎬

Tamil Content Creators

Tamil YouTube is one of India's largest regional language content ecosystems — channels covering cinema, politics, tech, and comedy reach millions. Creators use voice dictation for Tamil scripts, video descriptions, and social captions. Dictating in formal Tamil and editing for Tanglish is faster than writing from scratch.

📚

Students & Researchers

Students at Tamil-medium schools and universities use voice dictation for assignments and notes. Tamil language researchers — studying classical Sangam literature, inscriptions, and linguistic history — use dictation for academic writing. The formal centamil register needed for academic work matches what ASR models recognise best.

📰

Journalists & Writers

Tamil journalism — Dinamalar, Dinamani, The Hindu Tamil — reaches millions daily. Journalists and Tamil fiction writers use voice dictation for first drafts. Speaking in centamil and cleaning up the transcript is faster than composing formal Tamil prose on a keyboard.

🌍

Tamil Diaspora

Tamil communities in the UK (especially London), Canada (Toronto), Australia, Germany, and France use voice dictation for Tamil family communication, community organisation work, and Tamil language maintenance with children. Voice removes the friction of Tamil keyboard setup on non-Indian devices.

💼

IT Professionals

Chennai is one of India's largest IT hubs — Tamil-speaking tech professionals use voice dictation for Tamil-language internal communication, user documentation in Tamil, and personal notes. Heavy Tanglish code-switching is the norm; the Tamil portions dictated in ta-IN mode with English terms typed separately produces clean output.

தமிழ் குரல் கட்டளைகள் — Voice Commands in Tamil

Say these words during dictation to insert punctuation. Tamil traditionally uses the same punctuation marks as English in modern writing:

Punctuation / நிறுத்தற்குறிகள்

சொல்லுங்கள் / Say	சேர்க்கிறது / Inserts
"முற்றுப்புள்ளி"	. (full stop)
"காற்புள்ளி"	, (comma)
"கேள்விக்குறி"	?
"ஆச்சரியக்குறி"	!
"இருபுள்ளி"	:
"அரைப்புள்ளி"	;

Format / வடிவமைப்பு

சொல்லுங்கள் / Say	செயல் / Action
"புதிய வரி"	New line
"புதிய பத்தி"	New paragraph
"அழி"	Delete last word

Tamil punctuation note

Modern Tamil writing uses standard Western punctuation marks (., ?, !, :) — not a unique script like Hindi's danda or Japanese's 。. Tamil command terms (முற்றுப்புள்ளி etc.) map to these Western marks. Support varies by browser — Chrome has best Tamil punctuation command recognition.

தமிழ் ஆடியோ கோப்புகளை transcribe செய்யுங்கள் — MP3, WAV, MP4

Upload Tamil audio recordings — interviews, lectures, podcasts, meetings. Pro plan handles files up to 5 hours with timestamps. / தமிழ் ஒலி பதிவுகளை பதிவேற்றி தமிழ் உரையைப் பெறுங்கள்.

Pro திட்டங்களைப் பாருங்கள் →

அடிக்கடி கேட்கப்படும் கேள்விகள் — FAQ

Why does Tamil voice recognition make so many errors when I speak naturally?

Because natural spoken Tamil (koduntamil) is phonologically very different from the written Tamil (centamil) that the ASR model was trained on. When you say "போறேன்" (spoken "I'm going"), the model was trained on "போகிறேன்" (written form). The verb ending sounds different, the vowels differ, and the rhythm is different. This is Tamil's diglossia problem — it affects Tamil ASR more than almost any other language. The solution is to dictate in centamil: use full verb forms as written, even though it's not how you naturally speak.

Does it work for Sri Lankan Tamil (Jaffna)?

Poorly in natural Jaffna Tamil. Sri Lankan Tamil has preserved phonological distinctions (geminate consonants, archaic vowel length patterns) that Indian Tamil has lost, and its formal register differs from Indian centamil. There is no dedicated Sri Lankan Tamil locale. Jaffna Tamil speakers get significantly better results by approximating Indian centamil — using Indian Tamil verb forms and avoiding Sri Lanka-specific vocabulary. The acoustic differences between Jaffna Tamil and Indian centamil are significant enough that full accuracy is not achievable with current models.

தமிழ் குரல் உரை மாற்றம் Android மற்றும் iPhone-ல் வேலை செய்யுமா?

ஆம். Android-ல் Chrome மற்றும் iPhone-ல் Safari-ல் வேலை செய்கிறது. Android Chrome தமிழுக்கு சிறந்த முடிவுகளை தருகிறது. எந்த app-ஐயும் நிறுவ தேவையில்லை — நேரடியாக browser-ல் வேலை செய்கிறது. WhatsApp-ல் தமிழ் செய்திகளை அனுப்புவதற்கு: dictate செய்து, copy செய்து, WhatsApp-ல் paste செய்யுங்கள் — தமிழ் எழுத்துக்கள் சரியாக காட்டப்படும்.

What is the ழ (zha) sound and how do I produce it correctly for speech recognition?

The ழ (romanised "zh") represents a retroflex approximant /ɻ/ — a sound unique to Tamil and Malayalam. The tongue tip curls back and approximates (without touching) the post-alveolar region, producing a sound that is neither /r/ nor /l/. To produce it: start with the /r/ position, curl the tongue further back, and relax the contact. The word "தமிழ்" ends with this sound. If the model outputs ல or ர for words containing ழ, your ழ production is being interpreted as /l/ or /r/. Practicing the retroflex position improves recognition of the approximately 400+ Tamil words containing this phoneme.

Can I dictate in Tanglish (Tamil-English mix)?

Yes, with limitations. English words will be transliterated into Tamil script — "meeting" → "மீட்டிங்," "file" → "ஃபைல்." Tamil case markers attach correctly to English words: "meeting-ல," "file-ஐ." For formal content where English should remain in Roman script, pause dictation and type English terms manually. For WhatsApp, casual notes, and internal communication, Tamil-script English transliteration is serviceable and fast to clean up.

என் குரல் பதிவு செய்யப்பட்டு சேமிக்கப்படுகிறதா?

இல்லை. இலவச dictation கருவி Web Speech API ஐ பயன்படுத்துகிறது, இது browser-ல் built-in ஆக உள்ளது — உங்கள் குரல் எங்கள் servers-க்கு அனுப்பப்படவில்லை. Pro file upload feature-ல், processing க்காக மட்டும் file server-க்கு அனுப்பப்படும், transcription முடிந்தவுடன் தானாகவே delete ஆகும். எந்த ஒலிப் பதிவும் சேமிக்கப்படவில்லை.

Related Tools

🇮🇳

தமிழில் பேச ஆரம்பியுங்கள் — Start Dictating in Tamil

இலவசம், நிறுவல் இல்லை, பதிவு இல்லை. தமிழ் எழுத்துக்கள் தானாகவே வருகின்றன.

இப்போதே ஆரம்பிக்கவும் →

Chrome பரிந்துரைக்கப்படுகிறது — தமிழ் ASR-க்கு சிறந்த ஆதரவு