Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Speech Frameworkを使った音声認識の基本
Search
Masashi
January 29, 2025
Programming
0
73
Speech Frameworkを使った音声認識の基本
Mobile勉強会 ウォンテッドリー × チームラボ × Sansan #18 〜新技術の導入〜
https://sansan.connpass.com/event/338706/
Masashi
January 29, 2025
Tweet
Share
More Decks by Masashi
See All by Masashi
SpeechAnalyzerによるSpeech to Textの進化を探る
kawabe
0
8
Eight iOSを支えるアーキテクチャ
kawabe
1
620
これだけは伝えたい設計の技術
kawabe
0
1.3k
EightのUI Component化の取り組み
kawabe
0
140
Other Decks in Programming
See All in Programming
SUZURIの規約違反チェックにおけるクリエイタフィードバックの試⾏錯誤/Trial and Error in Creator Feedback for SUZURI's Terms of Service Violation Checks
ae14watanabe
1
150
詳細の決定を遅らせつつ実装を早くする
shimabox
1
1.1k
オフライン対応!Flutterアプリに全文検索エンジンを実装する @FlutterKaigi2025
itsmedreamwalker
2
210
乱雑なコードの整理から学ぶ設計の初歩
masuda220
PRO
31
12k
Rails Girls Sapporo 2ndの裏側―準備の日々から見えた、私が得たもの / SAPPORO ENGINEER BASE #11
lemonade_37
2
160
予防に勝る防御なし(2025年版) - 堅牢なコードを導く様々な設計のヒント / Growing Reliable Code PHP Conference Fukuoka 2025
twada
PRO
37
12k
「10分以内に機能を消せる状態」 の実現のためにやっていること
togishima
1
360
Atomics APIを知る / Understanding Atomics API
ssssota
1
150
AsyncSequenceとAsyncStreamのプロポーザルを全部読む!!
s_shimotori
1
280
Designing Repeatable Edits: The Architecture of . in Vim
satorunooshie
0
380
Java_プロセスのメモリ監視の落とし穴_NMT_で見抜けない_glibc_キャッシュ問題_.pdf
ntt_dsol_java
0
190
AI駆動開発ライフサイクル(AI-DLC)のホワイトペーパーを解説
swxhariu5
0
930
Featured
See All Featured
GraphQLとの向き合い方2022年版
quramy
49
14k
The Art of Programming - Codeland 2020
erikaheidi
56
14k
A designer walks into a library…
pauljervisheath
210
24k
4 Signs Your Business is Dying
shpigford
186
22k
Product Roadmaps are Hard
iamctodd
PRO
55
12k
The Web Performance Landscape in 2024 [PerfNow 2024]
tammyeverts
11
930
Done Done
chrislema
186
16k
How STYLIGHT went responsive
nonsquared
100
5.9k
How to Think Like a Performance Engineer
csswizardry
28
2.3k
Stop Working from a Prison Cell
hatefulcrawdad
272
21k
Practical Tips for Bootstrapping Information Extraction Pipelines
honnibal
24
1.6k
Refactoring Trust on Your Teams (GOTO; Chicago 2020)
rmw
35
3.2k
Transcript
Speech FrameworkΛͬͨ Իೝࣝͷجຊ
Տล խ࢙ɹMasashi Kawabe NOT A HOTEL גࣜձࣾ Smart Home νʔϜ
ιϑτΣΞΤϯδχΞ
ࣨʹεΠονϦϞίϯͳ͘ɺͯ͢ Home Controller͔Βૢ࡞ɻ ੈքதɺͲ͜ͷNOT A HOTELʹߦͬͯɺ ·ΔͰࣗͷΑ͏ʹ໎͏͜ͱͳͯ͘͢ͷػ ثͷૢ࡞͕Ͱ͖·͢ɻ Home Controller
Speech Framework • iOS ඪ४ͷԻೝࣝ Framework • ԻࡁΈͷԻϑΝΠϧͷೝࣝ, ϚΠΫೖྗΛར༻ͨ͠ϦΞϧλΠϜͳ Իͷೝ͕ࣝՄೳ
• ຊޠαϙʔτ • iOS 10.0 Ҏ߱Ͱར༻Մೳ
Speech Framework ΛͬͨԻೝࣝͷ࠷খ࣮
Իೝࣝͷதؒ݁Ռͷऔಘ • SFSpeechRecognitionRequest ͷ shouldReportPartialResults ϓϩύςΟͰɺԻೝࣝͷதؒ݁ՌΛऔಘ͢Δɾऔಘ͠ͳ͍ͷ ੍ޚ͕Մೳ • σϑΥϧτ true
( தؒ݁ՌΛऔಘ͢Δ ) • a • Իೝࣝͷ࠷ऴతͳ݁Ռ͚͕ͩඞཁͳ߹ɺ shouldReportPartialResults Λfalse ʹઃఆ͢Δ
Իೝࣝͷதؒ݁Ռऔಘͷ༗ແʹΑΔڍಈͷҧ͍
ΦϯσόΠεͰͷԻೝࣝ • ϓϥΠόγʔΛߟྀͯ͠ɺΦϯσόΠεͰԻೝࣝΛͤ͞Δ͜ͱ͕Ͱ͖Δ • ωοτϫʔΫʹܨ͕ͣͱԻೝ͕ࣝՄೳ • ͨͩ͠ɺServer ϕʔεͷԻೝࣝͱൺֱͯ͠ਫ਼͕ߴ͘ͳ͍ • Server
ϕʔεͷԻೝࣝɺԻೝࣝͷ࠷େ࣌ؒͷ੍ݶ, Ұ͋ͨΓͷճ੍ݶ͕͋Δͱ͍͏σϝϦοτ͕͋Δ • SFSpeechRecognitionRequest ͷ requiresOnDeviceRecognition ϓϩύςΟͰɺ༗ޮԽͰ͖Δ • a
ݴޠϞσϧͷΧελϚΠζ • ԻೝࣝͷݴޠϞσϧΛΧελϚΠζ͢Δ͜ͱͰɺಛఆͷϢʔεέʔε ͚ʹೝࣝਫ਼Λ্͛Δ͜ͱ͕Ͱ͖Δ • iOS 17.0 Ҏ߱Ͱར༻Մೳ • ΧελϚΠζͨ͠ݴޠϞσϧΛར༻͢ΔʹɺΦϯσόΠεͰԻೝࣝ
ͤ͞Δඞཁ͕͋Δ • a
ΧελϚΠζͨ͠ݴޠϞσϧͷ࡞
ΧελϚΠζͨ͠ݴޠϞσϧͷ࡞ • PhraseCount ΦϒδΣΫτΛར༻͢Δ͜ͱͰɺਖ਼֬ͳϑϨʔζΛ ΧελϜϞσϧʹొ͢Δ͜ͱ͕Ͱ͖Δ • ಛఆͷϑϨʔζΛೝࣝͤ͘͢͢͞Δ •
ΧελϚΠζͨ͠ݴޠϞσϧͷ࡞ • ΞϓϦͰઐ༻ޠͳͲҰൠతͰͳ͍୯ޠΛ༻͢Δ߹ɺ༻ޠͷεϖϧ ͱൃԻͷϖΞΛఆٛ͠ɺΧελϜϞσϧʹొ͢Δ͜ͱ͕Ͱ͖Δ • ൃԻ X-SAMPA ܗࣜ •
ΧελϚΠζͨ͠ݴޠϞσϧͷར༻ • ࡞ͨ͠ΧελϜݴޠϞσϧ SFSpeechRecognitionRequest ͷ customizedLanguageModel ϓϩύςΟʹઃఆ͢Δ͜ͱͰར༻Մೳ •
ࢀߟ • Advances in Speech Recognition • https://developer.apple.com/videos/play/wwdc2019/256 • Customize
on-device speech recognition • https://developer.apple.com/videos/play/ wwdc2023/10101 • Recognizing speech in live audio • https://developer.apple.com/documentation/speech/ recognizing-speech-in-live-audio