TTS AI Research Engineer

Mindlogic

Mindlogic

Software Engineering, Data Science
Posted on Jul 24, 2025
๐Ÿ—จ๏ธ

TTS AI Research Engineer

ํšŒ์‚ฌ ์†Œ๊ฐœ

ํ•จ๊ป˜ ๋” ์ด์•ผ๊ธฐ ํ•˜๊ณ  ์‹ถ์€ AI๋ฅผ ๋งŒ๋“œ๋Š” ์‚ฌ๋žŒ๋“ค, ๋งˆ์ธ๋“œ๋กœ์ง์ž…๋‹ˆ๋‹ค.
[๋งˆ์ธ๋“œ๋กœ์ง์ด ๋งŒ๋“ค์–ด๋‚ธ ์„ฑ๊ณผ]
๊ตญ๋‚ด์™ธ ์œ ์ˆ˜์˜ ํˆฌ์ž์ž๋“ค๋กœ๋ถ€ํ„ฐ ๋ˆ„์  ํˆฌ์ž 150์–ต, ํƒ„ํƒ„ํ•œ ์žฌ๋ฌด๊ตฌ์กฐ
6๋…„ ์ด์ƒ ๋”ฅ๋Ÿฌ๋‹ ๊ธฐ๋ฐ˜ ์ฑ—๋ด‡ ์ƒ์šฉ ์„œ๋น„์Šค ์ œ๊ณต ์ค‘
๋…์ฐฝ์ ์ธ ํŽ˜๋ฅด์†Œ๋‚˜ ๊ทธ๋ผ์šด๋”ฉ ๋ฐ ์žฅ๊ธฐ๊ธฐ์–ต ๊ธฐ์ˆ ์— ๊ธฐ๋ฐ˜ํ•œ ์„ธ๊ณ„ ์ตœ๊ณ  ์ˆ˜์ค€์˜ ํŽ˜๋ฅด์†Œ๋‚˜ ์ฑ—๋ด‡ ์—”์ง„ ๋ณด์œ 
์„œ์šธ๋Œ€ํ•™๊ต, ์„œ๊ฐ•๋Œ€ํ•™๊ต, ์ˆ™๋ช…์—ฌ์ž๋Œ€ํ•™๊ต ๋“ฑ ๋‹ค์ˆ˜์˜ ๋Œ€ํ•™์— AI์„œ๋น„์Šค ์ œ๊ณต ์ค‘
ํฌ๋ธŒ์Šค ์„ ์ • ํ•œ๊ตญ์ธ์ด ์‚ฌ๋ž‘ํ•œ ๋ชจ๋ฐ”์ผ์•ฑ 17์œ„, ์†Œ์…œ๋ถ€๋ฌธ 6์œ„ ๋‹ฌ์„ฑ
๊ตฌ๊ธ€ ์–ด์‹œ์Šคํ„ดํŠธ ํŠธ๋ž˜ํ”ฝ ๊ธ€๋กœ๋ฒŒ Top 5 ๋‹ฌ์„ฑ

ํฌ์ง€์…˜ ์ •๋ณด

์ง๋ฌด: TTS AI Research Engineer
๊ณ ์šฉ ํ˜•ํƒœ: ์ •๊ทœ์ง

ํ•ฉ๋ฅ˜ ์—ฌ์ •

์„œ๋ฅ˜์ „ํ˜•
์ž์œ  ํ˜•์‹์˜ ์ด๋ ฅ์„œ / ํฌํŠธํด๋ฆฌ์˜ค, PDFํ˜•์‹ ์ œ์ถœ
TTS ๊ด€๋ จ ํ”„๋กœ์ ํŠธ ๋ฐ ์—ฐ๊ตฌ ๊ฒฝํ—˜ ์ƒ์„ธ ๊ธฐ์ˆ  ํ•„์ˆ˜
recruit@mindlogic.ai ์— ์ด๋ฉ”์ผ๋กœ ์ œ์ถœ
์ธํ„ฐ๋ทฐ์ „ํ˜•
1๏ธโƒฃ ์˜จ๋ผ์ธ ๊ธฐ์ˆ  ์ธํ„ฐ๋ทฐ (ํฌํŠธํด๋ฆฌ์˜ค ๋ฐœํ‘œ ํฌํ•จ)
2๏ธโƒฃ ๋Œ€๋ฉด Tech & Culture Fit ์ธํ„ฐ๋ทฐ

์ฃผ์š”์—…๋ฌด

๐ŸŽฏ ํ˜„์žฌ ํ•ด๊ฒฐํ•ด์•ผ ํ•  ํ•ต์‹ฌ ๊ณผ์ œ

์ž์—ฐ์Šค๋Ÿฌ์šด ์ธํ† ๋„ค์ด์…˜ ๊ฐœ์„ : ํŽ˜๋ฅด์†Œ๋‚˜๋ณ„ ๋งํˆฌ์™€ ์–ต์–‘์„ ์ •ํ™•ํžˆ ์žฌํ˜„ํ•˜๋Š” TTS ๋ชจ๋ธ ๊ฐœ๋ฐœ
๊ฐ์ • ํ‘œํ˜„ ๊ณ ๋„ํ™”: ์›ƒ์Œ์†Œ๋ฆฌ, ํ•œ์ˆจ, ๊ฐํƒ„์‚ฌ ๋“ฑ ์ž์—ฐ์Šค๋Ÿฌ์šด ๊ฐ์ • ํ‘œํ˜„์ด ๊ฐ€๋Šฅํ•œ TTS ๊ตฌํ˜„
์ฒซ ํ† ํฐ ์ง€์—ฐ ์ตœ์†Œํ™”: ์‹ค์‹œ๊ฐ„ ๋Œ€ํ™”๋ฅผ ์œ„ํ•œ ultra-low latency TTS ์‹œ์Šคํ…œ ๊ตฌ์ถ•

๐Ÿš€ ํ•ต์‹ฌ ์—ฐ๊ตฌ๊ฐœ๋ฐœ ์˜์—ญ

Realtime Conversational Voice Cloning: ๋Œ€ํ™” ์ƒํ™ฉ์— ์ตœ์ ํ™”๋œ ์‹ค์‹œ๊ฐ„ ์Œ์„ฑ ๋ณต์ œ ๊ธฐ์ˆ  ๊ฐœ๋ฐœ
ํŽ˜๋ฅด์†Œ๋‚˜ ๊ธฐ๋ฐ˜ Expressive TTS: ์บ๋ฆญํ„ฐ๋ณ„ ๊ณ ์œ ํ•œ ์Œ์„ฑ ์Šคํƒ€์ผ๊ณผ ๊ฐ์ •์„ ๋ฐ˜์˜ํ•œ ๊ฐœ์ธํ™” ์Œ์„ฑํ•ฉ์„ฑ ์—”์ง„
Neural Audio Codec ์ตœ์ ํ™”
TTS ๋ฐ์ดํ„ฐ ํŒŒ์ดํ”„๋ผ์ธ: ์Œ์„ฑ ๋ฐ์ดํ„ฐ ์ „์ฒ˜๋ฆฌ, ์ •์ œ, ์ฆ๊ฐ•์„ ํ†ตํ•œ ๋ชจ๋ธ ์„ฑ๋Šฅ ํ–ฅ์ƒ

์ž๊ฒฉ์š”๊ฑด

์ปดํ“จํ„ฐ ๊ณตํ•™, ์ „๊ธฐ์ „์ž๊ณตํ•™, ๋˜๋Š” ๊ด€๋ จ ๋ถ„์•ผ ์„์‚ฌ ์ด์ƒ ๋˜๋Š” ์ด์— ์ค€ํ•˜๋Š” ์‹ค๋ฌด ๊ฒฝํ—˜ ๋ณด์œ 
TTS/์Œ์„ฑํ•ฉ์„ฑ ์—ฐ๊ตฌ๊ฐœ๋ฐœ ๊ฒฝํ—˜ 3๋…„ ์ด์ƒ
PyTorch, TensorFlow ๋“ฑ ๋”ฅ๋Ÿฌ๋‹ ํ”„๋ ˆ์ž„์›Œํฌ์— ๋Œ€ํ•œ ๊นŠ์€ ์ดํ•ด์™€ ํ™œ์šฉ ๊ฒฝํ—˜
์ตœ์‹  ๋”ฅ๋Ÿฌ๋‹ ๊ธฐ๋ฐ˜ TTS ์•Œ๊ณ ๋ฆฌ์ฆ˜ (FastSpeech, VITS, XTTS ๋“ฑ) ๊ตฌํ˜„ ๋ฐ ์ปค์Šคํ„ฐ๋งˆ์ด์ง• ๊ฒฝํ—˜
์Œ์„ฑ์‹ ํ˜ธ์ฒ˜๋ฆฌ ๊ธฐ์ดˆ ์ง€์‹: FFT, STFT, Mel-spectrogram, MFCC ๋“ฑ์˜ ์ดํ•ด์™€ ํ™œ์šฉ
TTS ๋ชจ๋ธ ํ•™์Šต ํŒŒ์ดํ”„๋ผ์ธ ๊ตฌ์ถ• ๊ฒฝํ—˜: ๋ฐ์ดํ„ฐ ์ „์ฒ˜๋ฆฌ, ํ•™์Šต, ์ถ”๋ก  ๋ฐ ํŠœ๋‹ ์ „๋ฐ˜
Python ๋ฐ ๊ด€๋ จ ์˜ค๋””์˜ค ์ฒ˜๋ฆฌ ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ (librosa, torchaudio ๋“ฑ) ์ˆ™๋ จ๋„
์˜์–ด ๊ธฐ์ˆ  ๋ฌธ์„œ ์ดํ•ด ๋ฐ ์ž‘์„ฑ ๊ฐ€๋Šฅํ•œ ์ˆ˜์ค€์˜ ์–ธ์–ด ๋Šฅ๋ ฅ

์šฐ๋Œ€์‚ฌํ•ญ

์‹ค์‹œ๊ฐ„ ๋Œ€ํ™”ํ˜• TTS ๊ตฌํ˜„ ๊ฒฝํ—˜ (ํŠนํžˆ ์ฒซ ํ† ํฐ ์ง€์—ฐ ์ตœ์†Œํ™”)
Emotional & Expressive TTS: ์›ƒ์Œ, ํ•œ์ˆจ, ๊ฐํƒ„์‚ฌ ๋“ฑ ์ž์—ฐ์Šค๋Ÿฌ์šด ๊ฐ์ • ํ‘œํ˜„ ๊ตฌํ˜„ ๊ฒฝํ—˜
Voice Cloning ๋ฐ Conversational TTS ๊ฐœ๋ฐœ ๊ฒฝํ—˜
์Œ์„ฑํ•ฉ์„ฑ ๊ด€๋ จ ๊ตญ์ œํ•™ํšŒ ๋…ผ๋ฌธ ๋ฐœํ‘œ: Interspeech, ICASSP, NeurIPS, ICLR ๋“ฑ
์ตœ์‹  TTS ๋ชจ๋ธ ์‹คํ—˜ ๊ฒฝํ—˜: VITS, XTTS, NeuralSpeech, SpeechT5, Bark, CSM ๋“ฑ
Neural Vocoder ์ตœ์ ํ™”: WaveNet, WaveGlow, HiFi-GAN, BigVGAN ๋“ฑ ์‹ค์‹œ๊ฐ„ ์ธํผ๋Ÿฐ์Šค ๊ตฌํ˜„ ๊ฒฝํ—˜
Neural Audio Codec ๋ชจ๋ธ ์‹คํ—˜ ๋ฐ ์ตœ์ ํ™” ๊ฒฝํ—˜: SNAC, Soundstream, encodec ๋“ฑ
์Œ์„ฑํ•™(Phonetics) ๋˜๋Š” ์–ธ์–ดํ•™ ๋ฐฐ๊ฒฝ์ง€์‹ (์ธํ† ๋„ค์ด์…˜ ํŒจํ„ด ์ดํ•ด)
TTS ์ƒ์šฉ ์„œ๋น„์Šค ์ ์šฉ ๋ฐ ์šด์˜ ๊ฒฝํ—˜ (API ์„œ๋ฒ„ ๊ตฌ์ถ•, ๋ฐฐํฌ ๋“ฑ)
MLOps ๋ฐ ๋ชจ๋ธ ์„œ๋น™ ๊ฒฝํ—˜ (Docker, Kubernetes, ํด๋ผ์šฐ๋“œ ์„œ๋น„์Šค)

๊ธฐ์ˆ ์Šคํƒ

ํ”„๋กœ๊ทธ๋ž˜๋ฐ ์–ธ์–ด: Python, TypeScript/JavaScript
๋”ฅ๋Ÿฌ๋‹ ํ”„๋ ˆ์ž„์›Œํฌ: PyTorch, TensorFlow, Hugging Face Transformers
๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค: PostgreSQL, Redis
ํด๋ผ์šฐ๋“œ ์„œ๋น„์Šค: AWS
์ปจํ…Œ์ด๋„ˆ ์˜ค์ผ€์ŠคํŠธ๋ ˆ์ด์…˜: Docker
CI/CD: GitHub Actions
๋ฒ„์ „ ๊ด€๋ฆฌ: Git
ํ˜‘์—… ๋„๊ตฌ: Slack, Jira, Notion
AI ๋„๊ตฌ: ChatGPT, Claude, Cursor

๊ทผ๋ฌดํ™˜๊ฒฝ ๋ฐ ๋ณต์ง€

์ฃผ 5์ผ ๊ทผ๋ฌด
์ž์œจ ์‹œ์ฐจ ์ถœํ‡ด๊ทผ (์‚ฐ์—…๊ธฐ๋Šฅ์š”์›/์ „๋ฌธ์—ฐ๊ตฌ์š”์›์€ ๋ณ‘๋ฌด์ฒญ์—์„œ ํ—ˆ๊ฐ€ํ•˜๋Š” ์œ ์—ฐ๊ทผ๋ฌด์ œ)
๊ฐ•๋‚จ๊ตฌ ์„ ์ •๋ฆ‰์—ญ ๋„๋ณด 1๋ถ„ ๋ฏธ๋งŒ ๊ฑฐ๋ฆฌ์˜ ๋‹จ๋… ์˜คํ”ผ์Šค
์ตœ์‹  ์—…๋ฌด์šฉ ๊ฐœ์ธ ๋งฅ๋ถ ์ œ๊ณต (๋งฅ๋ถM4)
๊ณ ์„ฑ๋Šฅ GPU ์„œ๋ฒ„ ์ง€์› (์Œ์„ฑ ๋ชจ๋ธ ํ•™์Šต์šฉ)
ChatGPT Pro or Claude / Cursor ๊ตฌ๋… ์ง€์›
๋™๋ฃŒ๋“ค์˜ ์ƒ์ผ ์ถ•ํ•˜ & ์„ ๋ฌผ
๊ทธ๋ฃน ์•กํ‹ฐ๋น„ํ‹ฐ ์„œํฌํŠธ
์ž์œจ๋ณต์žฅ

์ง€์›์‹œ ์ฐธ๊ณ ์‚ฌํ•ญ

์ง€์›์„œ ๋‚ด์šฉ, ๋˜๋Š” ์ „ํ˜• ์ง„ํ–‰ ์ค‘ ํ—ˆ์œ„ ์‚ฌ์‹ค์ด ์žˆ๋Š” ๊ฒฝ์šฐ ์ „ํ˜• ์ง„ํ–‰์ด ์ทจ์†Œ๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค
์ทจ์—…๋ณดํ˜ธ๋Œ€์ƒ์ž๋Š” ๊ด€๋ จ ๋ฒ•๊ทœ์— ์˜๊ฑฐํ•˜์—ฌ ์šฐ๋Œ€ํ•ฉ๋‹ˆ๋‹ค
์—ฐ๋ฝ์ฒ˜: recruit@mindlogic.ai
์ง€์› ๋งˆ๊ฐ: ์ˆ˜์‹œ ์ฑ„์šฉ (์šฐ์ˆ˜ ์ธ์žฌ ์ฑ„์šฉ ์‹œ ์กฐ๊ธฐ ๋งˆ๊ฐ ๊ฐ€๋Šฅ)