「怒鳴り声を穏やかにするAI」に本気で怒鳴ってみた　カスハラ対策で話題の技術、どんな声色になる？

Post LinkedIn

🗾閱讀原文： ITmedia AI+ (日本)

#voice-modulation#kasshara#de-escalation怒鳴り声穏やかaisoftbank

💡See how AI tames real shouts for harassment-proof service – audio AI breakthrough

⚡ 30 秒速覽

有什麼變化

AI 將叫喊轉為平靜語音防範騷擾

為什麼重要

可大幅降低服務業職場壓力，透過自動緩解自動化。提升即時音訊處理在安全領域的 AI 採用。

下一步行動

Test SoftBank's voice AI demo to integrate into your customer support pipeline.

誰應關注：Enterprise & Security Teams

關鍵要點

•AI 將叫喊轉為平靜語音防範騷擾
•與 SoftBank 合作真實測試大喊麥克風
•針對客戶服務濫用（カスハラ）情境

🧠 深度解析

背景與延伸：來自公開資料，非原文內容。引用 7 個來源。

🔑 增強重點摘要

•SoftBank's emotion-canceling AI was developed over three years and trained on 10,000+ voice data samples from 10 actors recording 100+ common phrases including screams, accusations, threats, and apology demands[1][2]
•The two-stage system first identifies angry voices and extracts key speech points, then uses acoustic tools to transform intonation into natural, polite tones while preserving the original words[1][2]
•The technology was inspired by a television program highlighting verbal abuse against call center staff, with developer Toshiyuki Nakatani specifically motivated to protect workers from customer harassment[2]

📊 競品分析▸ Show

Feature	SoftBank Emotion Canceling	Voicekiller	ElevenLabs Calm Voice Changer	YourVoic
Primary Use Case	Call center harassment mitigation	Creative voice direction/TTS	General voice transformation	Emotional text-to-speech
Training Data	10,000+ voice samples (anger-focused)	Not specified	Not specified	1000+ voices across 93+ languages
Emotional Control	Anger-to-calm conversion	Ultra-precise emotion control via instructions	Calm tone customization	Emotion-driven speech synthesis
Word Preservation	Yes (intonation only)	No (full speech control)	Yes	Yes
Implementation Status	Development phase (timeline unclear)[1]	Commercial product	Commercial product	Commercial product

🛠️ 技術深入

Two-stage architecture: Stage 1 uses AI voice-processing to identify angry callers and analyze speech characteristics; Stage 2 incorporates acoustic features of non-threatening voices to create calmer tones[2]
Training methodology: 10 actors recorded 100+ common phrases expressing various emotions (anger, frustration, accusations, threats, apology demands) generating 10,000+ total voice data samples[1][2]
Intonation modification: System significantly softens intonation while preserving original words and maintaining traces of anger so operators understand the situation[2]
Acoustic processing: Uses acoustic tools to transform aggressive speech into natural, even polite tones[1]

🔮 前景展望基於引用來源的 AI 分析

Workplace mental health applications beyond call centers

The emotion-canceling technology could extend to other high-stress customer-facing roles (retail, hospitality, emergency services) where verbal abuse impacts worker wellbeing.

Potential regulatory scrutiny over voice modification ethics

Altering customer voices without explicit consent may raise privacy and transparency concerns as the technology moves toward broader deployment.

Integration with broader AI harassment prevention ecosystems

SoftBank's system could combine with call recording, sentiment analysis, and automated escalation tools to create comprehensive customer service protection frameworks.