• 検索結果がありません。

分散型音声認識の商用システム構築

N/A
N/A
Protected

Academic year: 2021

シェア "分散型音声認識の商用システム構築"

Copied!
6
0
0

読み込み中.... (全文を見る)

全文

(1)2006−SLP−63(8)   2006/10/20. 社団法人 情報処理学会 研究報告 IPSJ SIG Technical Report. ಽᢔဳ㖸ჿ⹺⼂ߩ໡↪ࠪࠬ࠹ࡓ᭴▽ ട⮮ ᕡᄦ̐. ᴡ੗ ᕡ̐. ቝㇺች ᩕੑ̑. ̐ᩣᑼળ␠ KDDI ⎇ⓥᚲ ‫ޥ‬356-8502 ၯ₹⋵߰ߓߺ㊁Ꮢᄢේ 2-1-15 ̑KDDI ᩣᑼળ␠ ‫ޥ‬102-8460 ᧲੩ㇺජઍ↰඙㘵↰ᯅ 3-10-10 E-mail:. ̐{tkato, Hisashi.Kawai}@kddilabs.jp, ̑[email protected]. ޽ࠄ߹ߒ ៤Ꮺ㔚⹤ࠕࡊ࡝ࠤ࡯࡚ࠪࡦߩᣣᧄ⺆౉ജࠍᡰេߔࠆߚ߼‫ޔ‬2006 ᐕ 1 ᦬ࠃࠅࠦࡦࠪࡘ࡯ࡑะߌߦಽᢔဳ 㖸ჿ⹺⼂ߩࠨ࡯ࡆࠬࠍ㐿ᆎߒߚ‫ޕ‬៤Ꮺ㔚⹤ࡑࠗࠢߦ౉ജߐࠇߚ㖸ჿߪ៤Ꮺ㔚⹤ᯏ਄ߢ㖸㗀․ᓽ㊂ߦᄌ឵ߐࠇ‫ࠤࡄޔ‬ ࠶࠻ㅢାߢ㖸ჿ⹺⼂ࠨ࡯ࡃߦㅍାߐࠇࠆ‫ޕ‬៤Ꮺ㔚⹤߇㖸ჿ⹺⼂ࠨ࡯ࡃ߆ࠄฃାߒߚ⹺⼂⚿ᨐߪ↹㕙⴫␜ߐࠇࠆߚ߼‫ޔ‬ ࡙࡯ࠩߪ⍍ᤨߦ⹺⼂⚿ᨐࠍ⏕⹺ߒ‫ߚߞ⺋ࠍ⼂⹺ޔ‬႐วߦ߽⺋⹺⼂▎ᚲࠍㇱಽ⊛ߦୃᱜߔࠆߎߣ߇ߢ߈ࠆ‫ޕ‬㖸ჿ⹺⼂ ߦኻߔࠆࠬ࠻࡟ࠬ߿ਇ቟ࠍシᷫߔࠆߚ߼‫ޔ‬៤Ꮺ㔚⹤ᯏ਄ߩ․ᓽ㊂᛽಴ಣℂࠍ࡝ࠕ࡞࠲ࠗࡓൻߒᔕ╵ᤨ㑆ࠍᢙ⑽ߦ⍴ ❗ߔࠆߣߣ߽ߦ‫ߩ⼂⹺⺋ޔ‬น⢻ᕈ߇㜞޿႐วߦ‫ޟ‬ჿ߇ᄢ߈ߔ߉߹ߔ‫ޔޠ‬ ‫ޟ‬㔀㖸߇ᄢ߈ߔ߉߹ߔ‫ޠ‬ ‫ޔ‬ ‫⊒ޟ‬ჿ߇ᣧߔ߉߹ߔ‫ޠ‬ ߣ 3 ⒳㘃ߩࠕ࡜࡯ࡓࠍ⊒↢ߔࠆᯏ⢻ࠍㅊടߒߚ‫ߦ࠷ࡦ࠹ࡦࠦߩࠢ࡯ࡢ࠻࠶ࡀޔߚ߹ޕ‬ᣣ‫ޘ‬ㅊടߐࠇࠆᣂߒ޿ࠠ࡯ࡢ ࡯࠼ࠍ⹺⼂ߢ߈ࠆࠃ߁ߦߔࠆߚ߼‫ࠍࠬࡆ࡯ࠨޔ‬஗ᱛߖߕߦන⺆ㄉᦠ࡮ᢥᴺࠍᦝᣂߔࠆᯏ⢻ࠍ㐿⊒ߒߚ‫ޕ‬ ࠠ࡯ࡢ࡯࠼ ಽᢔဳ㖸ჿ⹺⼂, ៤Ꮺ㔚⹤. Development of a Commercial System of Distributed Speech Recognition Tsuneo KATO̐. Hisashi KAWAI̐. ̐KDDI R&D Laboratories Inc.. and. Eiji UTSUNOMIYȂ. 2-1-15 Ohara, Fujimino-shi, Satitama, 356-8502 Japan. ̑KDDI Corporation 3-10-10 Iidabashi, Chiyoda-ku, Tokyo, 102-8460 Japan E-mail:. ̐{tkato, Hisashi.Kawai}@kddilabs.jp, ̑[email protected]. Abstract To assist Japanese text input for applications on cellphones, a distributed speech recognition service for consumer applications was launched in January 2006. Speech input to a microphone is processed for acoustic feature extraction on the cellphone, then the features are transmitted to a speech recognition server by packet exchange, and recognition results received from the server are displayed on the screen. The recognition results are confirmed by sight, and partial correction of misrecognized words is possible if any. To reduce stress and unfamiliarity to speech recognition technology, response time from the server was shorten to a few seconds by real-time acoustic feature extraction on the cellphones, and warning function of three alarms, “Voice too loud ”, “Noise too loud”, and “Uttered too early”, were added to the client software. Moreover, a function of reloading new grammars and lexicons through a nonstop operation is equipped on the speech recognition server to enable frequent update of grammars and lexicons for network contents. Keyword Distributed Speech Recognition, Cellphone. 1. ߪ ߓ ߼ ߦ. ⹺⼂ࠍታⵝߒ‫ޔ‬ ‫ ޟ‬ჿ de ౉ ജ ‫ޠ‬ᯏ ⢻ ߣ ߒ ߡ 2006 ᐕ 1 ᦬ ߦ ࠦࡦࠪࡘ࡯ࡑะߌߩ㖸ჿ⹺⼂ࠨ࡯ࡆࠬࠍ㐿ᆎߒߚ‫ޕ‬ ᓥ᧪ߩ㔚⹤㖸ჿ⹺⼂ߣߪ⇣ߥࠅ‫ޔ‬៤Ꮺ㔚⹤ߩࡑࠗࠢ. ㄭ ᐕ ‫ޔ‬៤ Ꮺ 㔚 ⹤ ߪ 㔚 ⹤ ᯏ ߣ ߒ ߡ ߛ ߌ ߢ ߥ ߊ ‫ޔ ࡞ ࡯ ࡔޔ‬ ࠙ࠚࡉࡉ࡜࠙࠭‫▤࡞࡯ࡘࠫࠤࠬޔ‬ℂ╬߇ⴕ߃ࠆ៤Ꮺᖱ. ߆ࠄ౉ജߐࠇߚ㖸ჿߪ៤Ꮺ㔚⹤ᯏ਄ߢ㖸㗀․ᓽ㊂ߦᄌ. ႎ┵ᧃߣߒߡᐢߊ೑↪ߐࠇߡ޿ࠆ‫࡞࡯ࡔޕ‬૞ᚑ߿࠙ࠚ. ឵ߐࠇ‫ޔ‬㖸㗀․ᓽ㊂ߪࡄࠤ࠶࠻ㅢାߢ㖸ჿ⹺⼂ࠨ࡯ࡃ. ࡉ ߦ ߅ ߌ ࠆ URL ߿ ᬌ ⚝ ࠠ ࡯ ࡢ ࡯ ࠼ ߩ ౉ ജ ‫ޔ‬㔚 ⹤ Ꮽ ߿ ࠬ. ߦㅍାߐࠇࠆ‫ޕ‬㖸ჿ⹺⼂ࠨ࡯ࡃߩ⹺⼂⚿ᨐߪหߓߊࡄ. ࠤ ࠫ ࡘ ࡯ ࡞ Ꮽ ߩ ✬ 㓸 ߥ ߤ 10 ࠠ ࡯ ߦ ࠃ ࠆ ࠹ ࠠ ࠬ ࠻ ౉ ജ. ࠤ ࠶ ࠻ ㅢ ା ߢ ៤ Ꮺ 㔚 ⹤ ᯏ ߦ ㄰ ළ ߐ ࠇ ‫ ↹ޔ‬㕙 ⴫ ␜ ߐ ࠇ ࠆ ‫ޕ‬. ߪᣣᏱ⊛ߦ⊒↢ߒߡ޿ࠆ‫౉࠻ࠬࠠ࠹ޕ‬ജߩ⽶⩄ࠍシᷫ. 㔚⹤㖸ჿ⹺⼂ߩࠃ߁ߦ⹺⼂⚿ᨐࠍ㖸ჿߢ⏕⹺ߔࠆઍࠊ. ߔࠆߚ߼ߦ‫੍ޔ‬᷹ᄌ឵ᯏ⢻ࠍਛᔃߣߒߚᣣᧄ⺆౉ജࠪ. ࠅߦ↹㕙਄ߢ⏕⹺ߢ߈ࠆߚ߼‫ߚߞ⺋ࠍ⼂⹺ޔ‬႐วߢ߽. ࠬ ࠹ ࡓ ߩ ᡷ ⦟ ߇ ㅴ ࠎ ߢ ޿ ࠆ ߇ ‫ޔ‬10 ࠠ ࡯ ߩ ᠲ ૞ ߦ ᾘ ࠊ ߒ. ࡙࡯ࠩߪ⺋ߞߚ▎ᚲࠍ⍍ᤨߦ⷗ಽߌ‫ޔ‬ㇱಽ⊛ߦୃᱜߔ. ߐࠍᗵߓࠆ࡙࡯ࠩߪᄙ޿‫࠻ࠬࠠ࠹ߩࠩ࡯࡙ߚߒ߁ߎޕ‬. ࠆߎߣ߇ߢ߈ࠆ‫ޕ‬ ߹ߚ‫▚⸘ޔ‬㊂߿ᶖ⾌ࡔࡕ࡝ߩᄙ޿⹺⼂ಣℂࠍࠨ࡯ࡃ. ౉ ജ ࠍ ᡰ េ ߔ ࠆ ߚ ߼ ‫ ޔ‬KDDI ߪ ៤ Ꮺ 㔚 ⹤ ߦ ಽ ᢔ ဳ 㖸 ჿ. −39−.

(2) • 䇼⊒㚞䇽 {㚞} 䈎䉌 䇼⌕㚞䇽 䌻㚞 } {䉁䈪}. ߢታⴕߔࠆߚ߼‫ޔ‬៤Ꮺ㔚⹤ߩ࡝࠰࡯ࠬߦ೙㒢ߐࠇࠆߎ ߣߥߊᄢ⺆ᒵㅪ⛯㖸ჿ⹺⼂ࠍឭଏߔࠆߎߣ߇ߢ߈ࠆ‫ޕ‬ ߐࠄߦන⺆ㄉᦠߣᢥᴺ߇ࠨ࡯ࡃߦ⟎߆ࠇࠆߚ߼‫ࠚ࠙ޔ‬. • 䇼⊒㚞䇽 {㚞} 䈎䉌 䇼⌕㚞䇽 䌻㚞 } {䉁䈪}. ੹ ੹䈎䉌 䈜䈓䈮 ੹䈜䈓䈮 䈖䉏䈎䉌. • 䇼⊒㚞䇽 {㚞} 䈎䉌 䇼⌕㚞䇽 䌻㚞 } {䉁䈪}. ੹ᣣ ᣿ᣣ ᣿ᓟᣣ 䂾᦬䂾ᣣ. ࡉ╬ߩࡀ࠶࠻ࡢ࡯ࠢࠦࡦ࠹ࡦ࠷ߣㅪ៤ߒ߿ߔ޿‫ࡦࠦޕ‬ ࠹ࡦ࠷ߩᄌൻߦ޽ࠊߖߡන⺆ㄉᦠߣᢥᴺࠍᦝᣂߔࠆߎ ߣ߽ኈᤃߢ޽ࠆ‫ޕ‬. ಴⊒ ೔⌕. {䈱} 䂾ᤨ. 䂾ಽ 䈤䉊䈉䈬 ඨ. 䈮 㗃 䈎䉌 䉁䈪䈮. ಴⊒ ೔⌕. ໡↪ࠪࠬ࠹ࡓߩ᭴▽ߦ޽ߚࠅ‫ޔ‬㖸ჿ⹺⼂ߦኻߔࠆࠬ ࠻࡟ࠬ߿ਇᘠࠇߥ࡙࡯ࠩ߇ᛴߊਇ቟ࠍシᷫߔࠆߚ߼ߦ‫ޔ‬ • 䇼⊒㚞䇽 {㚞} 䈎䉌 䇼⌕㚞䇽 䌻㚞 } {䉁䈪}. 䂾ಽᓟ 䂾ᤨ㑆ᓟ 䂾ᤨ㑆ඨᓟ. 䈮 䈱. • 䇼⊒㚞䇽 {㚞} 䈎䉌 䇼⌕㚞䇽 䌻㚞 } {䉁䈪}. ੹ᣣ ᣿ᣣ ᣿ᓟᣣ 䂾᦬䂾ᣣ. ᆎ⊒ ⚳㔚 ᦨ⚳. 㜞ㅦߥᔕ╵ᕈߣ࡙࡯ࠩࡈ࡟ࡦ࠼࡝࡯ߥࠗࡦ࠲ࡈࠚ࡯ࠬ. ಴⊒ ೔⌕. ߩ㐿⊒ߦ㊀ὐࠍ߅޿ߚ‫ޕ‬ ೨⠪ߦߟ޿ߡߪ‫ޔ‬៤Ꮺ㔚⹤ߣࠨ࡯ࡃ㑆ߩㅢାᤨ㑆ࠍ ฽߼ߡࠬ࠻࡟ࠬߩߥ޿ᔕ╵ᤨ㑆ࠍታ⃻ߔࠆᔅⷐ߇޽ࠆ‫ޕ‬. {䈱}. ߘߩߚ߼‫⊒ޔ‬ჿ⚳ੌᓟ߹߽ߥߊ៤Ꮺ㔚⹤߆ࠄࠨ࡯ࡃߦ 㖸㗀․ᓽ㊂߇ㅍାߐࠇࠆߎߣ‫ޔ‬㖸ჿ⹺⼂ࠨ࡯ࡃߩ⹺⼂. ࿑ 1 ਸ਼ ឵ ᬌ ⚝ ࠨ ࡯ ࡆ ࠬ ߢ ⹺ ⼂ น ⢻ ߥ ᢥ ဳ. ಣℂ߇ 2 ⑽એౝߦቢੌߔࠆߎߣ߇ᔅⷐ᧦ઙߦߥߞߚ‫ޕ‬ ᓟ⠪ߦߟ޿ߡߪ‫ޔ‬ᓥ᧪ߩ㖸ჿ⹺⼂ࠨ࡯ࡆࠬߢߪ⹺⼂. ߎߣ. ⚿ᨐ߇⺋ߞߚ႐ว߿⹺⼂⚿ᨐ߇ᓧࠄࠇߥ߆ߞߚ႐วߦ‫ޔ‬. ࡮ 㖸ჿ⹺⼂ࠍㆡ↪ߒߚ႐วߦㆡᒰߥ⺆ᒵࠨࠗ࠭ߣ. ߥߗ⹺⼂ࠍ⺋ߞߚߩ߆‫⚿⼂⹺ߗߥޔ‬ᨐ߇ᓧࠄࠇߥ߆ߞ. ቯဳ⊛ߥ⊒ჿࡄ࠲ࡦ߇ᓧࠄࠇࠆߎߣ. ߚߩ߆‫ߪࠩ࡯࡙ޔ‬ේ࿃ࠍផ᷹ߔࠆᖱႎߐ߃ਈ߃ࠄࠇߥ. ߢ޽ࠆ‫⚿ߩߘޕ‬ᨐ‫ޔ‬ਸ਼឵ᬌ⚝ࠨ࡯ࡆࠬߣ⋡⊛࿾ᬌ⚝ࠨ. ߆ߞߚ‫ޔ߼ߚߩߘޕ‬ਇㆡಾߥ⊒ჿࠍ➅ࠅ㄰ߒߚࠅ‫ޔ‬㖸. ࡯ࡆࠬ߇ᦨೋߩㆡ↪ࠨ࡯ࡆࠬߣߐࠇߚ‫ޕ‬. ▵ࠍ඙ಾߞߡ⊒ჿߔࠆߎߣߦࠃࠅළߞߡ⹺⼂ߐࠇߦߊ. ᧄⓂߢߪ‫ޔ‬ᰴ▵ߢᦨೋߩࠕࡊ࡝ࠤ࡯࡚ࠪࡦࠍ⚫੺ߒ‫ޔ‬. ߊߥߞߚࠅߔࠆߎߣ߇ᄙ߆ߞߚ‫⚿⼂⹺ޕ‬ᨐ߇ᱜߒ޿߆. ╙ 3 ▵ߢࠪࠬ࠹ࡓ᭴ᚑ‫ ╙ޔ‬4 ▵ߢ៤Ꮺ㔚⹤ᯏ߳ߩࠢ࡜. ߤ߁߆‫⚿⼂⹺ޔ‬ᨐ߇⺋ߞߚ႐วߩේ࿃‫⚿⼂⹺ޔ‬ᨐ߇ᓧ. ࠗࠕࡦ࠻ታⵝߣ⺋⹺⼂ේ࿃ߩផቯㅢ⍮ᯏ⢻‫ ╙ޔ‬5 ▵ߢ. ࠄࠇߥ߆ߞߚ႐วߩේ࿃‫ࠍޔ‬ᱜ⏕ߦ․ቯߔࠆߎߣߪ࿎. 㖸ჿ⹺⼂ࠨ࡯ࡃߩ․ᓽࠍㅀߴࠆ‫ޕ‬. 㔍ߢ޽ࠆ߇‫ߦࠩ࡯࡙ޔ‬ᖱႎ߇ㅢ⍮ߐࠇߥ޿⁁ᴫࠍᡷༀ ߔࠆߚ߼‫ߩ⼂⹺⺋ޔ‬น⢻ᕈ߇㜞޿႐วߦផ᷹ߐࠇࠆℂ. 2. ࠕ ࡊ ࡝ ࠤ ࡯ ࠪ ࡚ ࡦ. ↱ࠍ࡙࡯ࠩߦㅢ⍮ߔࠆᯏ⢻ࠍ㐿⊒ߒߚ‫ޕ‬ㅢ⍮ߔࠆౝኈ ߪ ‫ ޟޔ‬ჿ ߇ ᄢ ߈ ߔ ߉ ߹ ߔ ‫ ⊒ ޟޔޠ‬ჿ ߇ ᣧ ߔ ߉ ߹ ߔ ‫ ޟޔޠ‬⢛. 2.1. ਸ਼ ឵ ᬌ ⚝ 䉰䊷䊎䉴. ᥊㔀㖸߇ᄢ߈ߔ߉߹ߔ‫ ߩޠ‬3 ⒳㘃ߢ޽ࠆ‫ޕ‬. ਸ਼឵ᬌ⚝ࠨ࡯ࡆࠬߪ‫ޔ‬ᜰቯߐࠇߚ⊒㚞‫⌕ޔ‬㚞ߣ಴⊒. ೨ㅀߩߣ߅ࠅ‫ޔ‬ಽᢔဳ㖸ჿ⹺⼂ߪࡀ࠶࠻ࡢ࡯ࠢ਄ߩ. ᤨೞ߽ߒߊߪ೔⌕ᤨೞߦኻߒߡ㔚ゞߩਸ਼឵ᖱႎࠍឭଏ. ࠺࡯࠲ࡌ࡯ࠬᬌ⚝ߦㆡߒߡ޿ࠆ‫ߦࠬ࡯ࡌ࠲࡯࠺ޕ‬ᣣ‫ޘ‬. ߔ ࠆ ࠨ ࡯ ࡆ ࠬ ߢ ޽ ࠆ ‫ ޕ‬GUI ࡌ ࡯ ࠬ ߩ ᧦ ઙ ౉ ജ ↹ 㕙 ߦ ࠹. ⊓㍳ߐࠇࠆᣂߒ޿ࠠ࡯ࡢ࡯࠼ࠍ⹺⼂ߢ߈ࠆࠃ߁ߦߔࠆ. ࠠࠬ࠻ࠍ౉ജߒ‫ࠍࠬ࡯ࡌ࠲࡯࠺ࠆ޽ߦࡃ࡯ࠨޔ‬ᬌ⚝ߔ. ߦߪ‫ޔ‬㖸ჿ⹺⼂ࠨ࡯ࡃߩන⺆ㄉᦠߣᢥᴺ㧔એ㒠ㄉᦠߣ. ࠆ ࠢ ࡜ ࠗ ࠕ ࡦ ࠻࡮ࠨ ࡯ ࡃ ဳ ߩ ࠕ ࡊ ࡝ ࠤ ࡯ ࠪ ࡚ ࡦ ߢ ޽ ࠆ ‫ޕ‬. ๭߱㧕ߦ߽ᣂߒ޿ࠠ࡯ࡢ࡯࠼ࠍ⊓㍳ߔࠆᔅⷐ߇޽ࠆ‫ޕ‬. ᓥ ᧪ ߩ 10 ࠠ ࡯ ౉ ജ ߦ ࠃ ࠆ ਸ਼ ឵ ᬌ ⚝ ߩ ႐ ว ‫ ⊒ ޔ‬㚞 ‫ޔ‬. ߎࠇ߹ߢㄉᦠᦝᣂᤨߦߪ㖸ჿ⹺⼂ࠨ࡯ࡃࠍ஗ᱛߒߥߌ. ⌕㚞‫ޔ‬಴⊒ᤨೞ߽ߒߊߪ೔⌕ᤨೞࠍߘࠇߙࠇߩࠬࡠ࠶. ࠇ߫ߥ߆ߞߚ‫ޔߒ߆ߒޕ‬ㄉᦠᦝᣂߩߚ߮㖸ჿ⹺⼂ࠨ࡯. ࠻ ߦ ౉ ജ ߔ ࠆ ‫ޕ‬㖸 ჿ ౉ ജ ߢ ߪ‫ ⊒ ޣޟ‬㚞 ‫ ⌕ ޣࠄ ߆ޤ‬㚞 ‫߹ޤ‬. ࡆ ࠬ ࠍ ஗ ᱛ ߔ ࠆ ߩ ߢ ߪ ‫ޔ‬㗫 ❥ ߥ ᦝ ᣂ ߇ 㔍 ߒ ߊ ߥ ࠆ ߚ ߼ ‫ޔ‬. ߢ ٤ ᦬ ٤ ᣣ ߩ ٤ ᤨ ٤ ಽ ߦ‫ ޣ‬಴ ⊒ 㧛 ೔ ⌕ ‫ޔ ߦ ߁ ࠃ ߁ ޿ ߣޠޤ‬. ࠨ࡯ࡆࠬࠍ஗ᱛߖߕߦㄉᦠᦝᣂߔࠆᯏ⢻ࠍ㐿⊒ߒߚ‫ޕ‬. ⶄᢙߩᬌ⚝᧦ઙࠍ৻ᐲߩ⊒ჿߦ฽߼ࠇ߫‫ߩࠇߙࠇߘޔ‬. એ਄ߩ․ᓽࠍ߽ߟಽᢔဳ㖸ჿ⹺⼂ߩᦨೋߩࠕࡊ࡝. ࠬࡠ࠶࠻ߦ⹺⼂⚿ᨐ߇⴫␜ߐࠇࠆ‫ޕ‬ਸ਼឵ᬌ⚝ߢ⹺⼂น. ࠤ࡯࡚ࠪࡦߪ‫ߩߘޔ‬ଢ೑ߐࠍᐢߊ࡙࡯ࠩߦ⸷߃߆ߌࠄ. ⢻ ߥ ᢥ ဳ ࠍ ࿑ 1 ߦ ␜ ߔ ‫ޕ‬ᣣ ᤨ ᜰ ቯ ߩ ⋭ ⇛ ߦ ኻ ߒ ߡ ߪ‫⃻ ޟ‬. ࠇ ‫ޔ‬᥉ ෸ ߩ ଦ ㅴ ߦ ❬ ߇ ࠆ ߎ ߣ ࠍ ᧦ ઙ ߣ ߒ ߡ ㆬ ቯ ߐ ࠇ ߚ ‫ޕ‬. ࿷ᤨೞߦ಴⊒‫࠻࡞ࠜࡈ࠺߇ޠ‬ᬌ⚝᧦ઙߣߒߡ౉ജߐࠇ. ౕ૕⊛ߦߪ‫ޔ‬. ࠆ ߇ ‫ ߩ ₸ ⼂ ⹺ޔ‬ૐ ਅ ࠍ 㒐 ߋ ߚ ߼ ේ ೣ ߣ ߒ ߡ‫ ⊒ ޣޟ‬㚞 ‫߆ޤ‬. ࡮ ࡙࡯ࠩ߇ᄙ޿ߎߣ. ࠄ‫ ⌕ ޣ‬㚞 ‫ ޣߢ ߹ޤ‬ᣣ ᤨ ‫ ޣߦޤ‬಴ ⊒ 㧛 ೔ ⌕ ‫ ⺆ ߩޠޤ‬㗅 ߪ ࿕. ࡮  ᓥ ᧪ ߩ 10 ࠠ ࡯ ߦ ࠃ ࠆ ࠹ ࠠ ࠬ ࠻ ౉ ജ ߇ ᄙ ޿ ߎ ߣ. ቯߦߒߡ޿ࠆ‫⊒ߩߎޕ‬ჿࡄ࠲ࡦߪ࿑ 2 ߩߣ߅ࠅ↹㕙⴫. ࡮ ᕆ޿ߢ޿ߚࠅᱠ߈ߥ߇ࠄߢ૶↪ߐࠇࠆ႐วߦ‫ޔ‬. ␜ߐࠇࠆߩߢ‫ޔ‬ਇᘠࠇߥ࡙࡯ࠩߪ⊒ჿࡄ࠲ࡦࠍ⋡ߢ⏕. 10 ࠠ ࡯ ᠲ ૞ ߇ 㔍 ߒ ߊ ‫ ޔ‬㖸 ჿ ౉ ജ ߇ ㆬ ߫ ࠇ ߿ ߔ ޿. ⹺ߒߥ߇ࠄ⊒ჿߔࠆߎߣ߇ߢ߈ࠆ‫ޕ‬. −40−.

(3) Ḱ஻ਛ. 䂓૑ᚲ. ⊒ჿߒߡߊߛߐ޿. • {䇼ㇺ㆏ᐭ⋵䇽} 䇼Ꮢ඙↸᧛䇽 {䇼↸ฬ䇽} {䇼࿾⇟䇽}. =಴⊒㚞?߆ࠄ =೔⌕㚞?߹ߢ =Q᦬Qᣣ?ߩ =QᤨQಽ?ߦ =೔⌕಴⊒?. ߅ᓙߜਅߐ޿. ㍳㖸⚳ੌ. 㩂㩢㨻㩁㨺ߢਛᢿ. 㩂㩢㨻㩁㨺ߢਛᢿ. 䂓㔚⹤⇟ภ 䇼࿕ቯ㔚⹤⇟ภ䇽 • 䇼៤Ꮺ㔚⹤⇟ภ䇽 䇼PHS㔚⹤⇟ภ䇽 䇼IP㔚⹤⇟ภ䇽 䂓ᐫฬ䋯ᣉ⸳ฬ. ㍳㖸ਛ. • {䇼࿾ฬ䇽}. ಽᨆਛ. =಴⊒㚞?߆ࠄ =೔⌕㚞?߹ߢ =Q᦬Qᣣ?ߩ =QᤨQಽ?ߦ =೔⌕಴⊒?. 䈱 䈮䈅䉎. 䇼ᐫฬ䋯ᣉ⸳ฬ䇽. 䂓㚞ฬ䋯ⓨ᷼ฬ •. ㍳㖸⚳ੌ. ߅ᓙߜਅߐ޿. 㩂㩢㨻㩁㨺ߢਛᢿ. 㩂㩢㨻㩁㨺ߢਛᢿ. 䇼㚞ฬ䇽 {㚞} 䇼ⓨ᷼ฬ䇽. ࿑ 3 ⋡ ⊛ ࿾ ᬌ ⚝ ࠨ ࡯ ࡆ ࠬ ߢ ⹺ ⼂ น ⢻ ߥ ᢥ ဳ. ࿑ 2 ਸ਼ ឵ ᬌ ⚝ ߩ ⊒ ჿ ᤨ ߩ ↹ 㕙 ㆫ ⒖. HTTP䊒䊨䊃䉮䊦 㖸㗀․ᓽ㊂䈫ㄉᦠฬ. ฦࠬࡠ࠶࠻ߦ౉ജߐࠇߚ⹺⼂⚿ᨐ߇⺋ߞߡ޿ࠆ႐ ว ߪ ‫ޔ‬ㇱ ಽ ⊛ ߦ ୃ ᱜ ߔ ࠆ ߎ ߣ ߇ ߢ ߈ ࠆ ‫ ୃޕ‬ᱜ ߩ ᣇ ᴺ ߪ ‫ޔ‬ ࡊ ࡞ ࠳ ࠙ ࡦ ࡔ ࠾ ࡘ ࡯ ߦ ࠃ ࠆ ೎ ୥ ⵬ ߩ ㆬ ᛯ ‫ޔ‬10 ࠠ ࡯ ߦ ࠃ. ⹺⼂⚿ᨐ. XML 䉰䊷䊋. 㖸ჿ⹺⼂ 䉰䊷䊋. ࠆᣣᧄ⺆౉ജ‫ޔ‬㖸ჿౣ౉ജ‫ ߩޔ‬3 ⒳㘃ߢ޽ࠆ‫ޕ‬㖸ჿౣ ౉ജߦࠃࠆୃᱜߪ‫ޔ‬ฦࠬࡠ࠶࠻ߩ㓞ߦ⟎߆ࠇߚࡑࠗࠢ 䉮䊮䊁䊮䉿 䉡䉢䊑 䉰䊷䊋. ࡏ࠲ࡦࠍㆬᛯߒߡⴕ߁‫ޕ‬ᬌ⚝᧦ઙߩ౉ജᓟ‫ޔ‬ᬌ⚝ࡏ࠲ ࡦࠍㆬᛯߔࠆߣਸ਼឵ᖱႎ߇ᓧࠄࠇࠆ‫ޕ‬. 2.2. ⋡ ⊛ ࿾ ᬌ ⚝ 䉰䊷䊎䉴 ࿑ 4 ಽ ᢔ ဳ 㖸 ჿ ⹺ ⼂ ࠪ ࠬ ࠹ ࡓ ߩ ᭴ ᚑ ⋡⊛࿾ᬌ⚝ࠨ࡯ࡆࠬߪ‫ޔ‬૑ᚲ‫ޔ‬㔚⹤⇟ภ‫ޔ‬ᐫฬ㧛ᣉ ⸳ฬ‫ޔ‬㚞㧛ⓨ᷼ฬߢᜰቯߐࠇߚ႐ᚲߩ࿾࿑ࠍ⴫␜ߒ‫ޔ‬ GPS ߢ ข ᓧ ߒ ߚ ⃻ ࿷ ࿾ ߆ ࠄ ⋡ ⊛ ࿾ ߹ ߢ ߩ ㆏ 㗅 ࠍ ࠟ ࠗ ࠼. 3. ಽ ᢔ ဳ 㖸 ჿ ⹺ ⼂ ࠪ ࠬ ࠹ ࡓ ߩ ᭴ ᚑ. ߔࠆࠨ࡯ࡆࠬߢ޽ࠆ‫ޕ‬ਸ਼឵ᬌ⚝ࠨ࡯ࡆࠬߩ႐วߣห᭽ ߦ‫౉ߩ߳࠻࠶ࡠࠬޔ‬ജߦಽᢔဳ㖸ჿ⹺⼂ࠍ↪޿ࠆ‫ޕ‬ ⋡⊛࿾ᬌ⚝ࠨ࡯ࡆࠬߦ߅ߌࠆ૑ᚲ౉ജ‫ޔ‬㔚⹤⇟ภ౉ ജ‫ޔ‬ᐫฬ㧛ᣉ⸳ฬ౉ജ‫ޔ‬㚞㧛ⓨ᷼ฬ౉ജߪࡔ࠾ࡘ࡯߇ ಽ߆ࠇߡ޿ࠆ‫ޕ‬ฦࡔ࠾ࡘ࡯ߢ⹺⼂น⢻ߥᢥဳࠍ࿑ 3 ߦ ␜ߔ‫ޕ‬ᐫฬ㧛ᣉ⸳ฬ౉ജߢߪ‫ޔ‬න⁛ߩᐫฬ㧛ᣉ⸳ฬߦ ട ߃ ߡ ‫ ޣޟ‬࿾ ฬ 㧛 㚞 ฬ ‫ ޣ ߩ ޤ‬ᐫ ฬ 㧛 ᣉ ⸳ ฬ ‫ࡦ ࠲ ࡄ ߩ ޠޤ‬ ࠍ⹺⼂ߢ߈ࠆ‫ߡߞࠃߦ࡯ࡘ࠾ࡔޕ‬ㄉᦠࠍಾࠅᦧ߃ࠆߚ ߼ߩㄉᦠฬ߇‫ޔ‬㖸㗀․ᓽ㊂ߣߣ߽ߦࠨ࡯ࡃߦㅍାߐࠇ ࠆ‫ޕ‬หࠨ࡯ࡆࠬߢߪએਅߩ 4 ⒳㘃ߩㄉᦠ߇ಾࠅᦧ߃ߡ ૶↪ߐࠇࠆ‫ޕ‬ z. 㚞ฬ㧛ⓨ᷼ฬ. z. ో࿖૑ᚲ. z. 㔚⹤⇟ภ. z. ో࿖ߩਥⷐᣉ⸳ฬ㧛ᐫฬ. ಽᢔဳ㖸ჿ⹺⼂ࠪࠬ࠹ࡓߩ᭴ᚑࠍ࿑ 4 ߦ␜ߔ‫ޕ‬៤Ꮺ 㔚 ⹤ ‫ޔ‬XML ࠨ ࡯ ࡃ ‫ޔ‬㖸 ჿ ⹺ ⼂ ࠨ ࡯ ࡃ ߢ ᭴ ᚑ ߐ ࠇ ࠆ ‫ޕ‬ಽ ᢔဳ㖸ჿ⹺⼂ߩࠨ࡯ࡃ⟲ߪࠕࡊ࡝ࠤ࡯࡚ࠪࡦߩࠦࡦ࠹ ࡦ࠷࠙ࠚࡉࠨ࡯ࡃߣߪ೎ߩࠨ࡯ࡃࠍ૶↪ߒߡ޿ࠆ‫ޕ‬៤ Ꮺ㔚⹤ߩࡑࠗࠢߦ౉ജߐࠇߚ㖸ჿߪ៤Ꮺ㔚⹤ᯏ਄ߢ㖸 㗀 ․ ᓽ ㊂ ߦ ᄌ ឵ ߐ ࠇ ‫ޔ‬XML ࠨ ࡯ ࡃ ࠍ ࠥ ࡯ ࠻ ߦ ߒ ߡ 㖸 ჿ ⹺⼂ࠨ࡯ࡃߦㅍାߐࠇࠆ‫ޕ‬㖸ჿ⹺⼂ࠨ࡯ࡃߦࠃࠆ⹺⼂ ⚿ ᨐ ߪ XML ࠼ ࠠ ࡘ ࡔ ࡦ ࠻ ߽ ߒ ߊ ߪ ┵ ᧃ ⴫ ␜ ↪ ߩ HTML ࠼ ࠠ ࡘ ࡔ ࡦ ࠻ ߦ ട Ꮏ ߐ ࠇ ߡ ៤ Ꮺ 㔚 ⹤ ߦ ㄰ ළ ߐ ࠇ ࠆ‫ޕ‬ ᣢሽߩ࠙ࠚࡉࠦࡦ࠹ࡦ࠷߿࠙ࠚࡉࡌ࡯ࠬߩࠕࡊ࡝ ࠤ࡯࡚ࠪࡦߦ㖸ჿ⹺⼂ࠍㅊടߒ߿ߔ޿ࠃ߁ߦ‫ޔ‬៤Ꮺ㔚 ⹤ ߣ XMLࠨ ࡯ ࡃ 㑆 ߩ ㅢ ା ߪ HTTP [ 3] ࠍ ૶ ↪ ߒ ߡ ޿ ࠆ ‫ޕ‬ ៤ Ꮺ 㔚 ⹤ ߆ ࠄ XMLࠨ ࡯ ࡃ ߳ ߩ 㖸 ჿ ⹺ ⼂ ⷐ ᳞ ߪ POSTࡔ. −41−.

(4) ࠰ ࠶ ࠼ ߢ ‫ޔ‬㖸 㗀 ․ ᓽ ㊂ ߣ ࠕ ࡊ ࡝ ࠤ ࡯ ࠪ ࡚ ࡦ ฬ ‫ޔ‬ㄉ ᦠ ฬ ‫ޔ‬. ⚳⹤ᬌ಴ᓟ߹߽ߥߊ㖸㗀․ᓽ㊂߇ㅍାߐࠇࠆࠃ߁. 㖸 㗀 ․ ᓽ ㊂ ߩ ⒳ 㘃 ฬ ࠍ ฽ ߻ XML߇ ฽ ߹ ࠇ ߡ ޿ ࠆ ‫ޕ‬㖸 ჿ. ߦ‫․ޔ‬ᓽ㊂᛽಴ࡊࡠࠣ࡜ࡓߪᢛᢙṶ▚ൻ‫࠲ࠬࠫ࡟ޔ‬ᄌ. ⹺ ⼂ ࠨ ࡯ ࡃ ߩ ࠗ ࡦ ࠲ ࡈ ࠚ ࡯ ࠬ ߽ HTTPߢ ޽ ࠆ ‫ޕ‬CGIࡊ ࡠ. ᢙߩෳᾖല₸ൻ‫ᦨߩ࠼࡯ࠦޔ‬ㆡൻ╬ࠍⴕ޿ࡀࠗ࠹ࠖࡉ. ࠣ࡜ࡓ߇⿠േߒ‫ޔ‬㖸ჿ⹺⼂ⷐ᳞߆ࠄ㖸㗀․ᓽ㊂ࠍขࠅ. ጀ ߦ ታ ⵝ ߐ ࠇ ߚ ‫ ⚿ ߩ ߘ ޕ‬ᨐ ‫ ޔ‬ES201108‫ ޔ‬ES202050 ߣ. ಴ߒ‫ޔ‬หߓߊ㖸ჿ⹺⼂ⷐ᳞߆ࠄขࠅ಴ߒߚㄉᦠฬࠍᜰ. ߽ߦ࡝ࠕ࡞࠲ࠗࡓಣℂ߇ⴕࠊࠇࠆࠃ߁ߦߥߞߚ‫ޕ‬. ቯߒߡ⹺⼂ಣℂࠍⴕ߁‫ޕ‬㖸ჿ⹺⼂ࠨ࡯ࡃߩᔕ╵ߪ‫ࠕޔ‬. ES201108 ߩ ಣ ℂ ⽶ ⩄ ߪ ES202050 ߩ ⚂ 1/3 ߢ ޽ ࠆ ‫ޕ‬. ࡊ ࡝ ࠤ ࡯ ࠪ ࡚ ࡦ ߦ ᔕ ߓ ߡ XML࠼ ࠠ ࡘ ࡔ ࡦ ࠻ ߿ HTML࠼. 4.2. ⺋ ⹺ ⼂ ේ ࿃ 䈱ផ ቯ ㅢ ⍮ ᯏ ⢻. ࠠࡘࡔࡦ࠻ߩࡈࠜ࡯ࡑ࠶࠻ߢ㄰ළߐࠇࠆ‫ޕ‬ XML ࠨ ࡯ ࡃ ߪ ‫ ޔ‬෺ ᣇ ะ ߩ ㅢ ା ߦ ฽ ߹ ࠇ ࠆ XML ࠼ ࠠ. ⺋⹺⼂ߩน⢻ᕈ߇㜞޿႐วߦ‫ޔ‬ផቯߐࠇࠆේ࿃ࠍ࡙. ࡘࡔࡦ࠻ߩࡈࠖ࡞࠲ߣߒߡᯏ⢻ߔࠆ‫࡚ࠪ࡯ࠤ࡝ࡊࠕޕ‬. ࡯ࠩߦㅢ⍮ߔࠆᯏ⢻ߣߒߡ‫౉ࠢࠗࡑޔ‬ജାภߩࠝ࡯ࡃ. ࡦߦᔕߓߚ㖸ჿ⹺⼂ࠨ࡯ࡃߩᝄࠅಽߌ‫ޔ‬ਇᱜߥ㖸ჿ⹺. ࡯ࡈࡠ࡯ᬌ಴‫ޔ‬ㆊᄢߥ⢛᥊㔀㖸ߩᬌ಴‫⹤ޔ‬㗡ಾᢿߩᬌ. ⼂ⷐ᳞ߩឃ㒰‫ޔ‬៤Ꮺ㔚⹤ᯏ⒳ߦࠃࠆࡈࠜ࡯ࡑ࠶࠻ߩ㆑. ಴ߩ 3 ᯏ⢻ࠍㅊടߒߚ‫ޕ‬3 ᯏ⢻ߪ޿ߕࠇ߽㖸ჿ⹺⼂ಣ. ޿ߩๆ෼‫ޔ‬㖸ჿ⹺⼂ⷐ᳞ߦ฽߹ࠇࠆ⹺⼂ߦਇⷐߥⷐ⚛. ℂࠍᔅⷐߣߖߕ‫ߩ࠻ࡦࠕࠗ࡜ࠢޔ‬ಣℂߛߌߢ್ቯߐࠇ. ߩ㒰෰‫⚿⼂⹺ޔ‬ᨐࠍࠠ࡯ߣߔࠆ࠺࡯࠲ࡌ࡯ࠬߩᬌ⚝╬. ࠆ߽ߩߢ޽ࠆ‫ޕ‬ ߥ߅‫್ޔ‬ቯߦ↪޿ߚ 3 ⒳㘃ߩࡄ࡜ࡔ࡯࠲ߦߟ޿ߡ‫ޔ‬. ࠍⴕ߁‫ޕ‬ ᬌ ⚝ ⚿ ᨐ ࠍ ᓧ ࠆ ߦ ߪ ‫ޔ‬10 ࠠ ࡯ ߦ ࠃ ࠆ ࠹ ࠠ ࠬ ࠻ ౉ ജ ߣ. ޽ࠆ୯ࠍႺߦ⹺⼂₸߇ᕆỗߦૐਅߔࠆߣ޿ߞߚ᣿⏕ߥ. ห᭽ߦࠬࡠ࠶࠻ߦᩰ⚊ߐࠇߚᬌ⚝᧦ઙࠍࠦࡦ࠹ࡦ࠷࠙. 㑐ଥ߇ᓧࠄࠇߥ߆ߞߚߚ߼‫ޔ‬㑣୯ࠍ⿥߃ߚ႐วߦߪ㖸. ࠚࡉࠨ࡯ࡃߦㅍାߒ‫ߩࡃ࡯ࠨࡉࠚ࠙࠷ࡦ࠹ࡦࠦޔ‬ᔕ╵. ჿ⹺⼂⚿ᨐߦട߃ߡ‫ޟ‬ჿ߇ᄢ߈߆ߞߚߚ߼⹺⼂⚿ᨐ߇. ࠍฃାߔࠆ‫ޕ‬. ᱜ ߒ ߊ ߥ ޿ น ⢻ ᕈ ߇ ޽ ࠅ ߹ ߔ ‫ ޟޔޠ‬㔀 㖸 ߇ ᄙ ߆ ߞ ߚ ߚ ߼ ⹺⼂⚿ᨐ߇ᱜߒߊߥ޿น⢻ᕈ߇޽ࠅ߹ߔ‫↹ߦ߁ࠃߩޠ‬. 4. ៤ Ꮺ 㔚 ⹤ ᯏ ߳ ߩ ࠢ ࡜ ࠗ ࠕ ࡦ ࠻ ታ ⵝ. 㕙⴫␜ࠍⴕ߁ߎߣߣߒߚ‫ޕ‬. 4.1. ․ ᓽ ㊂ 䈱᛽ ಴. 4.2.1. ࡑ ࠗ ࠢ ౉ ജ ା ภ ߩ ࠝ ࡯ ࡃ ࡯ ࡈ ࡠ ࡯ ᬌ ಴. ࡑࠗࠢ߆ࠄ౉ജߐࠇߚ㖸ჿߪ៤Ꮺ㔚⹤ᯏ਄ߢ㖸㗀. ࡑ ࠗ ࠢ ౉ ജ ା ภ ߇ A/D ᄌ ឵ ߩ ᦨ ᄢ ࡟ ࡌ ࡞ ࠍ ⿥ ߃ ࠆ ߣ. ․ᓽ㊂ߦᄌ឵ߐࠇ‫ߦࡃ࡯ࠨޔ‬ㅍାߐࠇࠆ‫ޕ‬㖸㗀․ᓽ㊂. ࠺ࠖࠫ࠲࡞ାภߪᦨᄢ୯㧛ᦨዊ୯ߢࠢ࡝࠶ࡇࡦࠣߐࠇ. ߪ ETSI (European Telecommunication Standard Institute). ࠆ‫ߚߒ↢⊒߇࡯ࡠࡈ࡯ࡃ࡯ࠝޕ‬ାภߩ⍴ᤨ㑆๟ᵄᢙಽ. ൘ ๔ ߩ ᮡ Ḱ ᣇ ᑼ ES201108. [ 1]. ߣ ES202050. [ 2]. ࠍណ↪ߒߚ‫ޕ‬. ᨆߩ⚿ᨐߦߪᧄ᧪ሽ࿷ߒߥ޿㜞๟ᵄᚑಽ߇⃻ࠇ‫⼂⹺ޔ‬. ࠨ ࡦ ࡊ ࡝ ࡦ ࠣ ๟ ᵄ ᢙ ߪ 8kHzߢ ޽ ࠆ ‫ ޕ‬ES201108 ߪ ⢛ ᥊. ₸ߩૐਅࠍ᜗ߊ‫࡞ࡊࡦࠨߚߒࠣࡦࡇ࠶࡝ࠢޔߢߎߘޕ‬. 㔀 㖸 ߩ ᛥ ࿶ ಣ ℂ ࠍ ฽ ߹ ߕ MFCC㖸 㗀 ․ ᓽ ㊂ ࠍ ᛽ ಴ ߒ ‫ޔ‬. ࠍ ࠞ ࠙ ࡦ ࠻ ߒ ‫ޔ‬㑣 ୯ ࠍ ⿥ ߃ ߚ ࠄ ࠕ ࡜ ࡯ ࡓ ࠍ ⊒ ↢ ߐ ߖ ࠆ ‫ޕ‬. ࡌ ࠢ ࠻ ࡞ ㊂ ሶ ൻ ߦ ࠃ ࠅ ࠺ ࡯ ࠲ ࠍ ࿶ ❗ ߔ ࠆ ‫ ޕ‬ES202050. ್ቯߦ↪޿ߚࡄ࡜ࡔ࡯࠲ߪᰴᑼߩߣ߅ࠅߢ޽ࠆ‫ޕ‬. ߪ 㖸 㗀 ․ ᓽ ㊂ ᛽ ಴ ߩ ೨ Ბ ߦ Wienerࡈ ࠖ ࡞ ࠲ ߦ ࠃ ࠆ ⢛ ᥊. N OFS. 㔀㖸ᛥ࿶ಣℂࠍ฽߻ߚ߼‫ޔ‬⠴㔀㖸ᕈ߇ᒝൻߐࠇߡ޿ࠆ. max nOFS t

(5) 1d t dT. nOFS t

(6) ߪ. ඨ 㕙 ‫ ▚ ⸘ ޔ‬㊂ ߇ ᄙ ߊ ‫ ޔ‬CPUߦ ኻ ߔ ࠆ ⽶ ⩄ ߇ ᄢ ߈ ޿ ‫ߘ ޕ‬ ߎߢ‫ޔ‬⢛᥊㔀㖸ߩᄢ߈ߐߦᔕߓߡ 2 ⒳㘃ߩ㖸㗀․ᓽ㊂. t ⇟⋡ߩ․ᓽ㊂᛽಴ࡈ࡟࡯ࡓߦ߅ߌࠆࠝ. ࡯ࡃ࡯ࡈࡠ࡯ࠨࡦࡊ࡞ߩᢙ‫ޔ‬T ߪ⊒ჿߩࡈ࡟࡯ࡓᢙߢ. ࠍ૶޿ಽߌࠆߎߣߦߒߚ‫ޕ‬ 㖸ჿߩขࠅㄟߺ߇ᆎ߹ࠆߣ‫⊒ޔ‬ჿࠍଦߔࡊࡠࡦࡊ࠻. ޽ࠆߩߢ. N OFS ߪ ⊒ ჿ ో ૕ ߩ ᦨ ᄢ ୯ ߢ ޽ ࠆ ‫࡝ ࡊ ࡦ ࠨ ޕ‬. ⴫␜ߩ೨ߩᢙ⊖ࡒ࡝⑽㑆㧔࿑ 2 ߩᏀ਄ߩ↹㕙㧕ߢ⢛᥊ 㔀㖸ࠍขᓧߔࠆ‫ߩߎޕ‬⢛᥊㔀㖸ߩᐔဋࡄࡢ࡯ࠍၮḰߦ. ࡦ ࠣ ๟ ᵄ ᢙ 8kHz ߢ ‫ ․ ޔ‬ᓽ ㊂ ᛽ ಴ ࡈ ࡟ ࡯ ࡓ ߩ ᄢ ߈ ߐ ߪ. ߒߡ‫⋧ޔ‬ኻࡄࡢ࡯ߦࠃࠅ⊒⹤ᬌ಴ࠍⴕ߁ߣߣ߽ߦ‫ޔ‬2. 25ms ߢ ޽ ࠆ ߚ ߼ ‫ޔ‬. nOFS t

(7) ߪ. 200 એ ਅ ߩ ୯ ߢ ޽ ࠆ ‫ޕ‬. ⒳㘃ߩ․ᓽ㊂᛽಴ᣇᑼࠍಾࠅᦧ߃ࠆ‫ޕ‬⢛᥊㔀㖸ߩᐔဋ ࡄ ࡢ ࡯ ߇ 㑣 ୯ એ ਅ ߢ ߪ ES201108 ߇ ㆬ ᛯ ߐ ࠇ ‫ ޔ‬㑣 ୯ એ ਄ ߢ ߪ ES202050 ߇ ㆬ ᛯ ߐ ࠇ ࠆ ‫ ⹤ ⊒ ޕ‬ᬌ ಴ ᓟ ߩ ౉ ജ ା. 4.2.2. ㆊ ᄢ ߥ ⢛ ᥊ 㔀 㖸 ߩ ᬌ ಴ ៤Ꮺ㔚⹤ᯏߦታⵝߒߚ㖸㗀․ᓽ㊂᛽಴ㇱߢߪ⢛᥊. ภߦኻߒߡ㗅ᰴ㖸㗀․ᓽ㊂߇᛽಴ߐࠇࠆ‫⹤⚳ޕ‬ᬌ಴߽ ⊒⹤ᬌ಴ߣห᭽ߦ⢛᥊㔀㖸ࡄࡢ࡯ࠍၮḰߣߔࠆ⋧ኻࡄ ࡢ࡯ߦၮߠ޿ߡⴕࠊࠇࠆ‫߇⹤⚳ޕ‬ᬌ಴ߐࠇࠆߣ‫ޔ‬㖸㗀 ಽ ᨆ ಣ ℂ ߪ ⚳ ੌ ߒ ‫ޔ‬㖸 㗀 ․ ᓽ ㊂ ߇ ࠨ ࡯ ࡃ ߦ ㅍ ା ߐ ࠇ ࠆ ‫ޕ‬. 㔀㖸߇ᄢ߈޿႐วߦ㔀㖸ᛥ࿶ಣℂࠍ฽߻ಽᨆᣇᑼࠍ૶ ↪ߒ‫ޔ‬㖸ჿ⹺⼂ࠨ࡯ࡃߢߪ⢛᥊㔀㖸ࠍ⠨ᘦߒߚ㖸㗀ࡕ ࠺࡞ࠍ↪޿ߡࡄ࠲ࡦࡑ࠶࠴ࡦࠣࠍⴕߞߡ޿ࠆ߇‫ޔ‬⢛᥊. −42−.

(8) 㔀㖸ߩჇᄢߦ઻޿⹺⼂₸ߪૐਅߒߡߒ߹߁‫ޕ‬. 㖸ჿ⹺⼂䉰䊷䊋. ߘߎߢ‫⊒ޔ‬ჿࡊࡠࡦࡊ࠻ߩ೨ߩᢙ⊖ࡒ࡝⑽ߢขᓧߒ ߚ⢛᥊㔀㖸࡟ࡌ࡞ߦၮߠ߈‫ޔ‬2 ⒳㘃ߩಽᨆᣇᑼࠍಾࠅ ᦧ߃ࠆߣߣ߽ߦ‫ߦࠄߐޔ‬೎ߩ㑣୯ࠃࠅᄢ߈޿႐วߦߪ. CGI 䊒䊨䉫䊤䊛. ធ⛯ⷐ᳞. ഀᒰ▤ℂ 䊒䊨䉫䊤䊛. ⁁ᘒ⋙ⷞ. 㖸ჿ⹺⼂ 䊒䊨䉶䉴. ㆊᄢߥ⢛᥊㔀㖸ߩࠕ࡜࡯ࡓࠍ⊒↢ߔࠆ‫ޕ‬හߜ‫ޔ‬ ࡮. Pnoise d T switch ߩ ႐ ว ࡮ ࡮ ࡮ ES201108. ࡮. Pnoise ! T switch ߩ ႐ ว ࡮ ࡮ ࡮ ES202050. ¾. ߐࠄߦ. ធ⛯. 㖸ჿ⹺⼂ 䊒䊨䉶䉴. Pnoise ! T alarm ߩ ႐ ว. 㖸ჿ⹺⼂ 䊒䊨䉶䉴. ࡮࡮࡮ㆊ ᄢ ⢛ ᥊ 㔀 㖸 ࠕ ࡜ ࡯ ࡓ ߎߎߢ‫ޔ‬. Pnoise ߪ ⢛ ᥊ 㔀 㖸 ࡟ ࡌ ࡞ ‫ ޔ‬T switch ߪ ಽ ᨆ ᣇ ᑼ. ಾ ᦧ ↪ ߩ 㑣 ୯ ‫ ޔ‬T alarm ߪ ࠕ ࡜ ࡯ ࡓ ⊒ ↢ ↪ ߩ 㑣 ୯ ߢ ‫ ޔ‬ᄢ ዊ 㑐 ଥ ߪ T switch.  T alarm ߢ ޽ ࠆ ‫ޕ‬. ࿑ 5㩷 㖸 ჿ ⹺ ⼂ 䉰䊷䊋䈱᭴ ᚑ. 5. 㖸 ჿ ⹺ ⼂ ࠨ ࡯ ࡃ 5.1. 㖸 ჿ ⹺ ⼂ 䉰䊷䊋䉸䊐䊃䉡䉣䉝. 4.2.3. ⹤ 㗡 ಾ ᢿ ߩ ᬌ ಴ ⹤㗡ಾᢿߪ‫ޔ‬㖸ჿߩขࠅㄟߺ߇ᆎ߹ࠆએ೨ߦ࡙࡯ࠩ. ࿑ 5 ߦ㖸ჿ⹺⼂ࠨ࡯ࡃߩ᭴ᚑࠍ␜ߔ‫ޕ‬หᤨߦⶄᢙߩ. ߇⊒ჿࠍ㐿ᆎߒߚ႐วߦ‫⊒ޔ‬ჿߩవ㗡߇෼㍳ߐࠇߥ޿. ⹺⼂ⷐ᳞ࠍಣℂߢ߈ࠆࠃ߁ߦ‫ⶄޔ‬ᢙߩ㖸ჿ⹺⼂ࡊࡠ࠮. ߎߣࠍᜰߒ‫⊒ޔ‬ჿߩవ㗡߇ᰳ⪭ߒߡ޿ࠆߚ߼ߦ⹺⼂₸. ࠬ߇⿠േߒߡ޿ࠆ‫ޕ‬㖸ჿ⹺⼂ࡊࡠ࠮ࠬߩ⁁ᘒߪ㨧⿠േ. ߩૐਅࠍ᜗ߊ‫ޕ‬. ਛ㧛ಣℂᓙߜ㧛ಣℂਛ㨩ߩ 3 ⁁ᘒߢഀᒰ▤ℂࡊࡠࠣ࡜. ᱜߒߊ෼㍳ߐࠇߚ⊒ჿߪ‫⹤⊒ޔ‬ᬌ಴೨ߣ⚳⹤ᬌ಴ᓟ. ࡓ ߦ ࠃ ࠅ ▤ ℂ ߐ ࠇ ࠆ ‫ ޕ‬CGI ࡊ ࡠ ࠣ ࡜ ࡓ ߆ ࠄ ߩ ⹺ ⼂ ⷐ ᳞. ߦή㖸ჿ඙㑆߇޽ࠅ‫ߥ⊛⊒⓭ޔ‬⢛᥊㔀㖸߇ߥߌࠇ߫ਔ. ߦኻߒߡ‫ޔ‬ഀᒰ▤ℂࡊࡠࠣ࡜ࡓߪಣℂᓙߜߩࡊࡠ࠮ࠬ. ⠪ߩࡄࡢ࡯ߪᄢ߈ߊ⇣ߥࠄߥ޿‫⹤ޕ‬㗡ಾᢿ߇⿠߈ߚ⊒. ࠍഀࠅᒰߡ‫ޔ‬ഀࠅᒰߡߚ㖸ჿ⹺⼂ࡊࡠ࠮ࠬߩ⁁ᘒࠍಣ. ჿߪ‫߇⹤⊒ޔ‬ᬌ಴ߐࠇߚ႐วߢ߽⊒⹤ᬌ಴೨ߦή㖸ჿ. ℂ ਛ ߦ ᄌ ߃ ࠆ ‫ޕ‬㖸 ჿ ⹺ ⼂ ࡊ ࡠ ࠮ ࠬ ߇ CGI ࡊ ࡠ ࠣ ࡜ ࡓ ߦ. ඙㑆߇ߥ޿ߚ߼‫ޔ‬㖸ჿขࠅㄟߺ㐿ᆎ⋥ᓟߩࡄࡢ࡯߇‫ޔ‬. ⹺⼂⚿ᨐࠍ㄰ළߒಣℂ߇ቢੌߔࠆߣ‫ߩࠬ࠮ࡠࡊߩߘޔ‬. ⚳⹤ᬌ಴ᓟߩή㖸ჿ඙㑆ߩࡄࡢ࡯ࠃࠅ߽㜞ߊߥࠆߣ⠨. ⁁ᘒߪಣℂᓙߜߦᚯࠆ‫ޔߚ߹ޕ‬ഀᒰ▤ℂࡊࡠࠣ࡜ࡓߪ. ߃ࠄࠇࠆ‫ޔߢߎߘޕ‬㖸ჿขࠅㄟߺ㐿ᆎ⋥ᓟߩኻᢙࡄࡢ. 㖸ჿ⹺⼂ࡊࡠ࠮ࠬ߇ਇᗧߦ⚳ੌߒߚ႐วߦ‫ޔ‬㖸ჿ⹺⼂. ࡯ߣ⚳⹤ᬌ಴ᓟߩኻᢙࡄࡢ࡯ߩᏅಽߦኻߒߡ㑣୯ಣℂ. ࡊࡠ࠮ࠬࠍ⥄േ⊛ߦౣ⿠േߔࠆ‫ޕ‬. ࠍⴕ޿‫ޔ‬㑣୯એ਄ߩ႐วߦࠕ࡜࡯ࡓࠍ⊒↢ߐߖࠆ‫ޕ‬හ. 㖸 ჿ ⹺ ⼂ ࡊ ࡠ ࠮ ࠬ ߪ ‫ ޔ‬CFGᢥ ᴺ ߣ ᧁ ᭴ ㅧ න ⺆ ㄉ ᦠ ߦ ၮ ߠ ߊ ࡢ ࡦ ࡄ ࠬ ߩ ᤨ 㑆 ห ᦼ ࡆ ࡯ ࡓ ត ⚝ ེ [ 4] ߢ ޽ ࠆ ‫↵ ޕ‬. ߜ ࡮. ᅚ೎ߩࠦࡦ࠹ࠠࠬ࠻ଐሽߩㅪ⛯⏕₸ኒᐲಽᏓ㖸⚛. P 10 log10 micon ! T early ߩ ႐ ว Pendpo int. HMMߣ 㔀 㖸 ࡕ ࠺ ࡞ [ 5] ࠍ ૶ ↪ ߒ ߡ ޿ ࠆ ‫ޕ‬㖸 㗀 ࡌ ࠢ ࠻ ࡞ ߪ ‫ޔ‬ ES20110 ߣ ES202050 ߩ 㖸 㗀 ․ ᓽ ㊂ ߦ ၮ ߠ ߈ ‫ ޔ‬MFCC‫ޔ‬. ࡮࡮࡮⹤㗡ಾᢿࠕ࡜࡯ࡓ ߎߎߢ‫ޔ‬. Pmicon ߪ 㖸 ჿ ข ࠅ ㄟ ߺ 㐿 ᆎ ⋥ ᓟ ߩ ࡄ ࡢ ࡯ ‫ޔ‬. Pendpo int ߪ ⚳ ⹤ ᬌ ಴ ᓟ ߩ ή 㖸 ჿ ඙ 㑆 ߩ ࡄ ࡢ ࡯ ‫ ޔ‬T early ߪ 㑣୯ߢ޽ࠆ‫ޕ‬. Ǎ MFCC‫ ޔ‬Ǎ Ǎ MFCC‫ ޔ‬Ǎ power‫ ޔ‬Ǎ Ǎ powerߩ 38 ᰴ ర ࠍ ↪ ޿ ߡ ޿ ࠆ ‫ ⊒ ޕ‬ჿ න ૏ ߢ CMS [ 㧢 ] ࠍ ⴕ ߞ ߡ ޿ ࠆ ‫ޕ‬ ᔕ╵ᤨ㑆ߪᐔဋߢ⚂ 1 ⑽ߢ޽ࠆ‫ޕ‬. 5.2. 䉰䊷䊎䉴ή ஗ ᱛ ㄉ ᦠ ᦝ ᣂ ᯏ ⢻ HTTP ࠍ ↪ ޿ ߚ ಽ ᢔ ဳ 㖸 ჿ ⹺ ⼂ ߪ ‫ࡦ ࠹ ࡦ ࠦ ࡉ ࠚ ࠙ ޔ‬ ࠷߿࠙ࠚࡉࡌ࡯ࠬࠕࡊ࡝ࠤ࡯࡚ࠪࡦߦ߅ߌࠆᬌ⚝᧦ઙ ౉ജߦㆡߒߡ޿ࠆ‫ߦࠬ࡯ࡌ࠲࡯࠺ߩࠢ࡯ࡢ࠻࠶ࡀޕ‬ ᣣ‫ޘ‬ㅊടߐࠇࠆᣂߒ޿ࠠ࡯ࡢ࡯࠼ࠍ⹺⼂ߢ߈ࠆࠃ߁ߦ ߔࠆߦߪ‫ޔ‬㖸ჿ⹺⼂ߩㄉᦠ࡮ᢥᴺࠍ㗫❥ߦᦝᣂߔࠆߎ. −43−.

(9) [2] ETSI ES202 050 v1.1.1 STQ; distributed speech recognition; advanced front-end feature extraction algorithm; compression algorithms. 2002.. ߣ߇ᔅⷐߦߥࠆ‫ᦝޕ‬ᣂߩߚ߮ߦࠨ࡯ࡆࠬࠍ஗ᱛߔࠆߩ ߢߪ㗫❥ߥㄉᦠᦝᣂ߇ਇน⢻ߦߥࠆߚ߼‫ࠍࠬࡆ࡯ࠨޔ‬ ஗ᱛߖߕߦㄉᦠࠍᦝᣂߔࠆᯏ⢻ࠍ㐿⊒ߒߚ‫ޕ‬ 1 บߩࠨ࡯ࡃߦߪⶄᢙߩ㖸ჿ⹺⼂ࡊࡠ࠮ࠬ߇⿠േߒ‫ޔ‬ ഀᒰ▤ℂࡊࡠࠣ࡜ࡓ߇ฦࡊࡠ࠮ࠬߩ⁁ᘒࠍ▤ℂߒߡ޿ ࠆ‫ߩߎޕ‬ഀᒰ▤ℂࡊࡠࠣ࡜ࡓ߆ࠄ‫ޔ‬ಣℂᓙߜߩ㖸ჿ⹺. [3] R. Fielding, J. Gettys, J. Mogul, H. Frystyk, L. Masinter, P. Leach, T. Berners-Lee, "Hypertext Transfer Protocol -- HTTP/1.1, " IETF RFC 2616, June 1999. [4] ቟ ⮮ ᓆ ↵ , ࡝ ࠕ ࡞ ࠲ ࠗ ࡓ 㖸 ჿ ⹺ ⼂ , ( ␠ ) 㔚 ሶ ᖱ ႎ ㅢ ା ቇ ળ , 2003.. ⼂ࡊࡠ࠮ࠬߩ৻ߟࠍ⿠േਛߦᄌ߃ߡ‫⹥ޔ‬ᒰߔࠆ㖸ჿ⹺ ⼂ࡊࡠ࠮ࠬࠍᣂߒ޿න⺆ㄉᦠߣᢥᴺߢౣ⿠േߔࠆ‫ߎޕ‬. [5] T. Kato and T. Shimizu, "Noise-Robust Cellular Phone Speech Recognition Using CODEC-Adapted Speech and Noise Models," Proc. ICASSP, May 2002.. ࠇࠍ⿠േߒߡ޿ࠆ㖸ჿ⹺⼂ࡊࡠ࠮ࠬಽߛߌ㗅⇟ߦ➅ࠅ ㄰ߔߎߣߢ‫ࠍࠬࡆ࡯ࠨޔ‬஗ᱛߖߕߦㄉᦠࠍᦝᣂߔࠆߎ ߣ߇ߢ߈ࠆ‫ޕ‬ᣂߒ޿න⺆ㄉᦠߣᢥᴺߪ੍߼ᬌ⸽↪ࠨ࡯. [6] 㤥 ጤ , ട ⮮ , ᮘ ญ , "ᦨ ࠁ ߁ ⁁ ᘒ ♽ ೉ ࠍ ↪ ޿ ߚ ታ ᤨ 㑆 ࠤ ࡊ ࠬ ࠻ ࡜ ࡓ ᐔ ဋ ୯ ᱜ ⷙ ൻ ߩ ᬌ ⸛ ," 㔚 ሶ ᖱ ႎ ㅢ ା ቇ ળ ⺰ ᢥ ⹹ , Vol. J82-D2, No.3, pp.332-339, 1999.. ࡃߢേ૞⏕⹺ߐࠇ‫࡝ࡕࡔޔ‬਄ߩන⺆ㄉᦠߣᢥᴺ߇ࡃࠗ ࠽࡝ㄉᦠࡈࠔࠗ࡞ߣߒߡ಴ജߐࠇࠆ‫ޕ‬໡↪ࠨ࡯ࡃߪߘ ߩࡃࠗ࠽࡝ㄉᦠࡈࠔࠗ࡞ࠍ⺒ߺㄟ߻ߎߣߢ‫ޔ‬ㄉᦠᦝᣂ ᤨߩࠛ࡜࡯ࠍ࿁ㆱߒ‫ޔ‬ㄉᦠᦝᣂߩᤨ㑆ࠍᢙච⑽ߦᛥ߃ ࠆߎߣ߇ߢ߈ࠆ‫ޕ‬. 6. ߅ ࠊ ࠅ ߦ ៤Ꮺ㔚⹤ࠕࡊ࡝ߦ߅ߌࠆᣣᧄ⺆౉ജࠍᡰេߔࠆߚ ߼ߦ៤Ꮺ㔚⹤ߦታⵝߒߚಽᢔဳ㖸ჿ⹺⼂ࠪࠬ࠹ࡓࠍ⚫ ੺ߒߚ‫࡯ࠤ࡝ࡊࠕࠬ࡯ࡌࡉࠚ࠙߿࠷ࡦ࠹ࡦࠦࡉࠚ࠙ޕ‬ ࡚ࠪࡦߣߩⷫ๺ᕈࠍ㜞߼ࠆߚ߼‫ޔ‬៤Ꮺ㔚⹤࡮ࠨ࡯ࡃ㑆 ߩ ㅢ ା ߦ ߪ HTTP ࡊ ࡠ ࠻ ࠦ ࡞ ࠍ ↪ ޿ ‫ ޔ‬៤ Ꮺ 㔚 ⹤ ߆ ࠄ ࠨ ࡯ࡃ߳ߪ㖸㗀․ᓽ㊂ߦട߃ߡ‫ޔࡦ࡚ࠪ࡯ࠤ࡝ࡊࠕޔ‬ㄉ ᦠߩ⼂೎ሶࠍㅍାߔࠆ‫ޕ‬ 㖸ჿ⹺⼂ߦኻߔࠆࠬ࠻࡟ࠬ߿ਇ቟ࠍシᷫߔࠆߚ߼‫ޔ‬ ⹺⼂ᕈ⢻ߦട߃ߡ‫ޔ‬㜞ㅦߥᔕ╵ᕈߣ࡙࡯ࠩࡈ࡟ࡦ࠼࡝ ࡯ߥࠗࡦ࠲ࡈࠚ࡯ࠬߩ㐿⊒ߦ㊀ὐࠍ߅޿ߚ‫ޕ‬៤Ꮺ㔚⹤ ᯏ਄ߢߩ࡝ࠕ࡞࠲ࠗࡓ․ᓽ㊂᛽಴‫ޔ‬㖸ჿ⹺⼂ࠨ࡯ࡃߦ ߅ߌࠆᐔဋ 1 ⑽㑆ߩ⹺⼂ಣℂߦࠃࠅ‫⊒ޔ‬ჿ⚳ੌ߆ࠄ⹺ ⼂⚿ᨐߩขᓧ߹ߢߩᤨ㑆ࠍᢙ⑽ߦᛥ߃ߚ‫ߩ⼂⹺⺋ޕ‬႐ ว‫⚿⼂⹺ޔ‬ᨐ߇ߥ޿႐วߦߘߩℂ↱߇࡙࡯ࠩߦឭଏߐ ࠇ ߥ ޿ ⁁ ᴫ ࠍ ᡷ ༀ ߔ ࠆ ߚ ߼ ‫ ޟ‬ჿ ߇ ᄢ ߈ ߔ ߉ ߹ ߔ ‫ ޟޔޠ‬㔀 㖸 ߇ ᄢ ߈ ߔ ߉ ߹ ߔ ‫ ⊒ ޟޔޠ‬ჿ ߇ ᣧ ߔ ߉ ߹ ߔ ‫ ߩ ޠ‬3 ⒳ 㘃 ߩ ࠕ࡜࡯ࡓࠍ⊒↢ߐߖࠆ઀⚵ߺࠍࠢ࡜ࠗࠕࡦ࠻࠰ࡈ࠻࠙ ࠛࠕߦታⵝߒߚ‫ޕ‬ ߹ߚ‫ࡦ࡚ࠪ࡯ࠤ࡝ࡊࠕࡉࠚ࠙߿࠷ࡦ࠹ࡦࠦࡉࠚ࠙ޔ‬ ߩ࠺࡯࠲ࡌ࡯ࠬߦᣣ‫⊓ޘ‬㍳ߐࠇࠆᣂߒ޿ࠠ࡯ࡢ࡯࠼ࠍ ⹺⼂ߢ߈ࠆࠃ߁ߦࠨ࡯ࡆࠬࠍ஗ᱛߖߕߦන⺆ㄉᦠߣᢥ ᴺࠍᦝᣂߔࠆᯏ⢻ࠍ㐿⊒ߒߚ‫ޕ‬. ᢥ. ₂. [1] ETSI ES201 108 v.1.1.2 distributed speech recognition; front-end feature extraction algorithm; compression algorithm. 2000.. −44−.

(10)

参照

関連したドキュメント

In order to estimate the noise spectrum quickly and accurately, a detection method for a speech-absent frame and a speech-present frame by using a voice activity detector (VAD)

patient with apraxia of speech -A preliminary case report-, Annual Bulletin, RILP, Univ.. J.: Apraxia of speech in patients with Broca's aphasia ; A

For the rest of this paper, let A denote a K- algebra isomorphic to Mat d +1 (K) and let V denote an irreducible left A-module. It is helpful to think of these primitive idempotents

Unfortunately, the method fails if someone tries to use it for proving the left hand side of the Hermite–Hadamard- type inequality for a generalized 4-convex function since, by the

In the second computation, we use a fine equidistant grid within the isotropic borehole region and an optimal grid coarsening in the x direction in the outer, anisotropic,

Theorem 4.8 shows that the addition of the nonlocal term to local diffusion pro- duces similar early pattern results when compared to the pure local case considered in [33].. Lemma

After briefly summarizing basic notation, we present the convergence analysis of the modified Levenberg-Marquardt method in Section 2: Section 2.1 is devoted to its well-posedness

While conducting an experiment regarding fetal move- ments as a result of Pulsed Wave Doppler (PWD) ultrasound, [8] we encountered the severe artifacts in the acquired image2.