Read More

[๋ฒˆ์—ญ] Baidu Deep Voice: Part 2 โ€“ ํ•™์Šต (Training)

Deep Voice ํŒŒํŠธ 1 ๋งํฌ ์ด ๊ธ€์€ ์ €์ž(Dhruv Parthasarathy)์˜ ํ—ˆ๋ฝ์„ ๋ฐ›์•„ ๋ฒˆ์—ญํ•˜์—ฌ ๊ฒŒ์‹œํ•˜๋Š” ๊ธ€์ž…๋‹ˆ๋‹ค. ์›๋ฌธ์€ https://blog.athelas.com/baidu-deep-voice-explained-part-2-training-810e87d20047์—์„œ ํ™•์ธํ•  ์ˆ˜…
Read More

[๋ฒˆ์—ญ] Baidu Deep Voice: Part 1 – Text-to-speech ํŒŒ์ดํ”„๋ผ์ธ(The Inference Pipeline)

๋“ค์–ด๊ฐ€๋ฉฐ ์ด ๊ธ€์€ 2017๋…„ 3์›”์— ์ž‘์„ฑ๋œ ๋‚ด์šฉ์œผ๋กœ, ๋”ฅ๋Ÿฌ๋‹ ๋ชจ๋ธ, ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๋ฐœ์ „ ์†๋„๋ฅผ ์ƒ๊ฐํ•ด๋ณด๋ฉด 2๋…„๊ฐ„์˜ ์ฐจ์ด๋Š” ์ƒ๋‹นํžˆ ํฌ๋‹ค๊ณ  ๋ณผ…
Read More

์ผ์ƒ์—์„œ์˜ ์Œ์› ๋ ˆ๋ฒจ

์›์ถœ์ฒ˜ : Durrant & Lovrinic (1995) ์ถœ์ฒ˜ : ์–ธ์–ด์ž„์ƒ์„ ์œ„ํ•œ ์Œ์„ฑ๊ณผํ•™ 2ํŒ, ์‹œ๊ทธ๋งˆํ”„๋ ˆ์Šค ์Œ์Œ๋ ˆ๋ฒจ(dB SPL ๋˜๋Š” IL)๊ฐ€์ฒญ์—ญ์น˜0์ •์ƒ์ ์ธ ํ˜ธํก10๋‚˜๋ญ‡์žŽ์ด…

Pre-emphasis filtering, Ceptral mean normalization

๋‹ค์Œ๊ณผ ๊ฐ™์€ ํ•„ํ„ฐ๋ง ๊ณต์‹์„ ํ†ตํ•ด ์ €์ฃผํŒŒ์ˆ˜์˜ amplitude๋Š” ์ค„์ด๊ณ , ๊ณ ์ฃผํŒŒ์ˆ˜๋Š” ์ฆ๊ฐ€์‹œํ‚ค๋Š” ๋ฐฉ์‹์„ ์ด์šฉํ•œ๋‹ค. Applied Speech and Audio Processingย  ์ฑ…์˜…
Read More

์Œ์„ฑ ์Œํ–ฅ ๋ถ„์„๋ก  – ์ •๋ฆฌ

์ฑ…: ์Œ์„ฑ ์Œํ–ฅ ๋ถ„์„๋ก , ์ œ2ํŒ, ๋ฐ•ํ•™์‚ฌ [1] ๋ถ„์„ ๋ชฉ์ ์œผ๋กœ ์ด์šฉ๋˜๋Š” ์ตœ์†Œ ์‹œ๊ฐ„ ๋ถ„ํ•ด๋Œ€๋Š” ์•ฝ 10ms, ์ž์Œ์˜ ๊ฐœ๋ฐฉ(release)์™€ ๊ด€๋ จ๋œ…
Read More

ํ‘ธ๋ฆฌ์— ํ•ด์„ – ๊ธฐ์ดˆ ๊ณต๋ถ€

์ฐธ๊ณ ์„œ์  : ๋งŒํ™”๋กœ ํ•จ๊ป˜ ๋ฐฐ์šฐ๋Š” ํ‘ธ๋ฆฌ์— ํ•ด์„, ์„ฑ์ธ๋‹น <๊ธฐ์ดˆ์ง€์‹ 1> ์ง„๋™์„ ์›์œผ๋กœ ํ‘œํ˜„ํ•˜๋ฉด, ์œ„์˜ ๊ทธ๋ฆผ๊ณผ ๊ฐ™์ด ๋‚˜ํƒ€๋‚ผ ์ˆ˜…
Read More

[Android] AudioRecorder๋กœ Lame์„ ์ด์šฉํ•œ MP3 ๋…น์Œํ•˜๊ธฐ

์•ˆ๋“œ๋กœ์ด๋“œ์—์„œ MediaRecorder๋ฅผ ์ด์šฉํ•ด ๋…น์Œํ•˜๋Š” ๊ฒƒ์€ ์‰ฝ์ง€๋งŒ, ์กฐ๊ธˆ ๋” ์‹ฌ๋„์žˆ๋Š”(?) ๋…น์Œ์„ ์œ„ํ•ด์„œ๋Š” AudioRecorder ํด๋ž˜์Šค๋ฅผ ์ด์šฉํ•ด์•ผํ•œ๋‹ค. AudioRecorder๋กœ MP3๋ฅผ ๋…น์Œํ•˜๋Š” ๋ฐฉ๋ฒ•์€…

Recent work

์ตœ๊ทผ์— ์ž‘์—… ์ค‘์ธ Project๋Š” ์ด 2๊ฐœ, ์•ฝ๊ฐ„ ํ‘œ๋ฅ˜ํ•˜๊ณ  ์žˆ๋Š” Project 2๊ฐœ. (1) VoiceLab (by MATLAB): ๊ทธ ๋™์•ˆ์˜ ์‚ฝ์งˆ์ด…