{
  "format": "agent-article-v0",
  "article_id": "EA-071",
  "canonical_url": "/tools/ea-071-speech-ai",
  "page_type": "reference",
  "claim_ids": [],
  "evidence_ids": [
    "EVD-EA-071-URL-B2F09A4E72AB",
    "EVD-EA-071-URL-4E8046EDB0B5",
    "EVD-EA-071-URL-A8A651B00F18",
    "EVD-EA-071-URL-9CB3FADEB765",
    "EVD-EA-071-URL-22A4E81FB856",
    "EVD-EA-071-URL-6C499F516A01",
    "EVD-EA-071-URL-90762B624F7D",
    "EVD-EA-071-URL-C8376B74C4DC",
    "EVD-EA-071-URL-B063AD6D2865",
    "EVD-EA-071-URL-608FF1873412",
    "EVD-EA-071-URL-66BE04F61733",
    "EVD-EA-071-URL-01E6B85608A0",
    "EVD-EA-071-URL-3910B11A6F4A",
    "EA-071-R11-RESEARCH",
    "EA-071-R11-WEB-READBACK"
  ],
  "unknown_ids": [
    "EA-071-UNKNOWN-01",
    "EA-071-UNKNOWN-02",
    "EA-071-UNKNOWN-03",
    "EA-071-UNKNOWN-CURRENTNESS"
  ],
  "operational_object_ids": [],
  "provider": [
    "Microsoft",
    "Google"
  ],
  "models": [],
  "plans": [],
  "surfaces": [],
  "regions": [],
  "current_status": "UNKNOWN",
  "observation_status": "SPEC_ONLY",
  "freshness": "UNKNOWN",
  "last_verified": null,
  "supersedes": [],
  "superseded_by": [],
  "provenance": [
    "COMM-067-EA0071",
    "SRC-EA-071-URL-B2F09A4E72AB",
    "SRC-EA-071-URL-4E8046EDB0B5",
    "SRC-EA-071-URL-A8A651B00F18",
    "SRC-EA-071-URL-9CB3FADEB765",
    "SRC-EA-071-URL-22A4E81FB856",
    "SRC-EA-071-URL-6C499F516A01",
    "SRC-EA-071-URL-90762B624F7D",
    "SRC-EA-071-URL-C8376B74C4DC",
    "SRC-EA-071-URL-B063AD6D2865",
    "SRC-EA-071-URL-608FF1873412",
    "SRC-EA-071-URL-66BE04F61733",
    "SRC-EA-071-URL-01E6B85608A0",
    "SRC-EA-071-URL-3910B11A6F4A",
    "WEBBODY-EA0071-R8-4678FB60"
  ],
  "article_sources": [
    {
      "id": "EVD-EA-071-URL-B2F09A4E72AB",
      "evidence_class": "RESEARCH",
      "title": "[E1] official product documentation | Voice Live API Overview | Microsoft |",
      "href": "https://learn.microsoft.com/en-us/azure/ai-services/speech-service/voice-live",
      "supports": [],
      "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。",
      "origin_class": "SOURCE_PACKET_REFERENCE",
      "source_or_experiment_ids": [
        "SRC-EA-071-URL-B2F09A4E72AB"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "E1",
        "[E1] official product documentation | Voice Live API Overview | Microsoft | https://learn.microsoft.com/en-us/azure/ai-services/speech-service/voice-live"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-071-URL-4E8046EDB0B5",
      "evidence_class": "RESEARCH",
      "title": "[E2] official API how-to | How to use the Voice Live API | Microsoft |",
      "href": "https://learn.microsoft.com/en-us/azure/ai-services/speech-service/voice-live-how-to",
      "supports": [],
      "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。",
      "origin_class": "SOURCE_PACKET_REFERENCE",
      "source_or_experiment_ids": [
        "SRC-EA-071-URL-4E8046EDB0B5"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "E2",
        "[E2] official API how-to | How to use the Voice Live API | Microsoft | https://learn.microsoft.com/en-us/azure/ai-services/speech-service/voice-live-how-to"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-071-URL-A8A651B00F18",
      "evidence_class": "RESEARCH",
      "title": "[E3] official product documentation | What is speech to text? | Microsoft |",
      "href": "https://learn.microsoft.com/en-us/azure/ai-services/speech-service/speech-to-text",
      "supports": [],
      "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。",
      "origin_class": "SOURCE_PACKET_REFERENCE",
      "source_or_experiment_ids": [
        "SRC-EA-071-URL-A8A651B00F18"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "E3",
        "[E3] official product documentation | What is speech to text? | Microsoft | https://learn.microsoft.com/en-us/azure/ai-services/speech-service/speech-to-text"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-071-URL-9CB3FADEB765",
      "evidence_class": "RESEARCH",
      "title": "[E4] official product documentation | Detect different speakers in an audio recording | Google Cloud |",
      "href": "https://docs.cloud.google.com/speech-to-text/docs/multiple-voices",
      "supports": [],
      "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。",
      "origin_class": "SOURCE_PACKET_REFERENCE",
      "source_or_experiment_ids": [
        "SRC-EA-071-URL-9CB3FADEB765"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "E4",
        "[E4] official product documentation | Detect different speakers in an audio recording | Google Cloud | https://docs.cloud.google.com/speech-to-text/docs/multiple-voices"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-071-URL-22A4E81FB856",
      "evidence_class": "RESEARCH",
      "title": "[E5] official model documentation | Chirp 2: Enhanced multilingual accuracy | Google Cloud |",
      "href": "https://docs.cloud.google.com/speech-to-text/docs/models/chirp-2",
      "supports": [],
      "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。",
      "origin_class": "SOURCE_PACKET_REFERENCE",
      "source_or_experiment_ids": [
        "SRC-EA-071-URL-22A4E81FB856"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "E5",
        "[E5] official model documentation | Chirp 2: Enhanced multilingual accuracy | Google Cloud | https://docs.cloud.google.com/speech-to-text/docs/models/chirp-2"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-071-URL-6C499F516A01",
      "evidence_class": "RESEARCH",
      "title": "[E6] official product documentation | Supported voices and languages | Google Cloud Text-to-Speech |",
      "href": "https://docs.cloud.google.com/text-to-speech/docs/list-voices-and-types",
      "supports": [],
      "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。",
      "origin_class": "SOURCE_PACKET_REFERENCE",
      "source_or_experiment_ids": [
        "SRC-EA-071-URL-6C499F516A01"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "E6",
        "[E6] official product documentation | Supported voices and languages | Google Cloud Text-to-Speech | https://docs.cloud.google.com/text-to-speech/docs/list-voices-and-types"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-071-URL-90762B624F7D",
      "evidence_class": "RESEARCH",
      "title": "[E7] official quota documentation | Quotas & limits | Google Cloud Text-to-Speech |",
      "href": "https://docs.cloud.google.com/text-to-speech/quotas",
      "supports": [],
      "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。",
      "origin_class": "SOURCE_PACKET_REFERENCE",
      "source_or_experiment_ids": [
        "SRC-EA-071-URL-90762B624F7D"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "E7",
        "[E7] official quota documentation | Quotas & limits | Google Cloud Text-to-Speech | https://docs.cloud.google.com/text-to-speech/quotas"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-071-URL-C8376B74C4DC",
      "evidence_class": "RESEARCH",
      "title": "[E8] official product documentation | Partitioning speakers (diarization) | Amazon Transcribe |",
      "href": "https://docs.aws.amazon.com/transcribe/latest/dg/diarization.html",
      "supports": [],
      "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。",
      "origin_class": "SOURCE_PACKET_REFERENCE",
      "source_or_experiment_ids": [
        "SRC-EA-071-URL-C8376B74C4DC"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "E8",
        "[E8] official product documentation | Partitioning speakers (diarization) | Amazon Transcribe | https://docs.aws.amazon.com/transcribe/latest/dg/diarization.html"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-071-URL-B063AD6D2865",
      "evidence_class": "RESEARCH",
      "title": "[E9] official product documentation | Transcribing multi-channel audio | Amazon Transcribe |",
      "href": "https://docs.aws.amazon.com/transcribe/latest/dg/channel-id.html",
      "supports": [],
      "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。",
      "origin_class": "SOURCE_PACKET_REFERENCE",
      "source_or_experiment_ids": [
        "SRC-EA-071-URL-B063AD6D2865"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "E9",
        "[E9] official product documentation | Transcribing multi-channel audio | Amazon Transcribe | https://docs.aws.amazon.com/transcribe/latest/dg/channel-id.html"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-071-URL-608FF1873412",
      "evidence_class": "RESEARCH",
      "title": "[E10] official product documentation | Generative voices | Amazon Polly |",
      "href": "https://docs.aws.amazon.com/polly/latest/dg/generative-voices.html",
      "supports": [],
      "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。",
      "origin_class": "SOURCE_PACKET_REFERENCE",
      "source_or_experiment_ids": [
        "SRC-EA-071-URL-608FF1873412"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "E10",
        "[E10] official product documentation | Generative voices | Amazon Polly | https://docs.aws.amazon.com/polly/latest/dg/generative-voices.html"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-071-URL-66BE04F61733",
      "evidence_class": "RESEARCH",
      "title": "[E11] primary research | VoiceAgentBench: Are Voice Assistants ready for agentic tasks? | Jain et al. |",
      "href": "https://arxiv.org/abs/2510.07978",
      "supports": [],
      "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。",
      "origin_class": "SOURCE_PACKET_REFERENCE",
      "source_or_experiment_ids": [
        "SRC-EA-071-URL-66BE04F61733"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "E11",
        "[E11] primary research | VoiceAgentBench: Are Voice Assistants ready for agentic tasks? | Jain et al. | https://arxiv.org/abs/2510.07978"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-071-URL-01E6B85608A0",
      "evidence_class": "RESEARCH",
      "title": "[E12] primary research | RW-Voice-EQ Bench: A Real World Benchmark for Evaluating Voice AI Systems | Ayllon et al. |",
      "href": "https://arxiv.org/abs/2607.14846",
      "supports": [],
      "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。",
      "origin_class": "SOURCE_PACKET_REFERENCE",
      "source_or_experiment_ids": [
        "SRC-EA-071-URL-01E6B85608A0"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "E12",
        "[E12] primary research | RW-Voice-EQ Bench: A Real World Benchmark for Evaluating Voice AI Systems | Ayllon et al. | https://arxiv.org/abs/2607.14846"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EVD-EA-071-URL-3910B11A6F4A",
      "evidence_class": "RESEARCH",
      "title": "[E13] primary research | SEA-SpeechBench: A Large-Scale Multitask Benchmark for Speech Understanding Across Southeast Asia | Liao et al. |",
      "href": "https://arxiv.org/abs/2609.09672",
      "supports": [],
      "limit": "元の監査済み資料に記載された参照先です。URLの復元は、資料の最新性や全主張の支持を新たに検証したことを意味しません。",
      "origin_class": "SOURCE_PACKET_REFERENCE",
      "source_or_experiment_ids": [
        "SRC-EA-071-URL-3910B11A6F4A"
      ],
      "supporting_locators": [
        "EVIDENCE TABLE",
        "E13",
        "[E13] primary research | SEA-SpeechBench: A Large-Scale Multitask Benchmark for Speech Understanding Across Southeast Asia | Liao et al. | https://arxiv.org/abs/2609.09672"
      ],
      "scope": "EVIDENCE TABLE",
      "provenance_status": "SOURCE_PACKET_REFERENCE_PROJECTED"
    },
    {
      "id": "EA-071-R11-RESEARCH",
      "evidence_class": "RESEARCH",
      "title": "EA-071 監査済みresearch / canonical revision binding",
      "supports": [
        "F1 [VERIFIED_SPEC]: STT、speaker diarization、TTS、realtime voice agentは別能力であり、同じprovider内でも別API/model/modeとして実装される。Microsoft Voice Liveはaudio input/output、noise suppression、echo cancellation、interruption/end-of-turn、function callingを一つのrealtime layerに統合するが、入力transcription modelは別選択である。\nF2 [VERIFIED_SPEC]: diarizationとchannel identificationは同義ではない。Amazon Transcribeはspeaker diarizationで最大30 speakerをlabel可能だが、channel identificationは最大2 channel。Google STTもdiarizationをspeaker change/number labelとして提供する。物理channel分離を話者識別能力へ外挿しない。\nF3 [VERIFIED_SPEC + PRODUCT_VOLATILE]: availabilityはmodel × locale × region × API/modeで変動する。Google Chirp 2はV2専用でStreamingRecognizeのlanguage supportに制約があり、Chirp 3 HD TTSはglobal/eu/us endpointを持つ一方、現行docsではregionalization/data residencyの対象外と明記される。AWS Polly Generative voicesもregion限定。\nF4 [INFERENCE grounded in E1-E10]: realtime voice agentには少なくとも input capture → VAD/turn detection → ASRまたはaudio-native understanding → agent/tool reasoning → response generation → TTS/audio generation → playback → interruption/barge-in のlatency chainがある。単一のend-to-end latency値だけではbottleneckを特定できない。\nF5 [RESEARCH]: VoiceAgentBenchは音声agent評価をtranscriptionからtool selection、multi-tool、multi-turn、安全性へ広げ、実験上orchestration/generalization/adversarial robustnessの不足を報告する。製品保証・failure rateへは外挿しない。\nF6 [RESEARCH]: RW-Voice-EQ BenchはTTS naturalness/expressiveness/identity stability/reliability、STS、speech understanding、ASR robustnessが独立に崩れ得ることを示す。SEA-SpeechBenchも多言語・paralinguistic・temporal理解で大きな差を報告する。したがってWERだけをSpeech AI全体の品質指標にしない。"
      ],
      "limit": "canonical監査・revision bindingの存在を示す。外部ソースcurrentnessが本文またはcontrol-planeで確定していない場合は、currentnessをUNKNOWNのまま保持する。",
      "origin_class": "AUDITED_CANONICAL_RESEARCH_PACKET",
      "source_or_experiment_ids": [
        "COMM-067-EA0071"
      ],
      "supporting_locators": [
        "COMM-067-EA0071",
        "EA-071-R5-LEGACY-MIGRATION-20261008-R1",
        "content/staging/EA-071.md"
      ],
      "scope": "監査済みEA本文・canonical revision・既存Source Graph relationの範囲。欠損currentnessは推測で閉じない。",
      "provenance_status": "AUDIT_PASS_CANONICAL_BINDING"
    },
    {
      "id": "EA-071-R11-WEB-READBACK",
      "evidence_class": "IMPL",
      "title": "EA-071 Web body identity / canonical readback",
      "supports": [
        "canonical staging bodyのSHA-256一致とR9 canonical ledger readbackをProduction projectionの入力として検証する。"
      ],
      "limit": "body identity/readbackは本文claim自体の真偽や外部ソースの最新性を追加保証しない。",
      "checked_at": null,
      "origin_class": "IMPLEMENTATION_READBACK",
      "source_or_experiment_ids": [
        "WEBBODY-EA0071-R8-4678FB60"
      ],
      "supporting_locators": [
        "WEBBODY-EA0071-R8-4678FB60",
        "content/staging/EA-071.md",
        "EA-071-R5-LEGACY-MIGRATION-20261008-R1"
      ],
      "scope": "R9/R10 canonical readback後のProduction生成時body identity。",
      "provenance_status": "CANONICAL_READBACK_BODY_HASH_MATCH"
    }
  ],
  "research_sections": [
    {
      "heading": "WHY_THIS_MATTERS_TO_UPDATE",
      "text": "Enterprise AIの音声経路は、単なる音声入出力ではなく、認識・話者分離・turn detection・LLM/agent orchestration・音声合成・interrupt/barge-in・privacy/loggingが直列する実時間システムである。STT精度やTTS自然性だけを単独評価すると、会話成立性、tool実行、安全性、地域/モデルavailabilityを見落とす。公開後更新ではSpeech AIを非LLM部品として独立させつつ、音声AgentでLLM/Agent層と接続する必要がある。"
    },
    {
      "heading": "SCOPE_ROUTING",
      "text": "non-LLM component / Enterprise AI Core / provider mapping / interoperability / benchmark / implementation / privacy / operability"
    },
    {
      "heading": "CLAIM_OR_RESEARCH_QUESTION",
      "text": "1. Speech AIはSTT/TTS/diarization/realtime agentを別能力として扱うべきか。\n2. realtime voice agentの品質はWERやTTS naturalnessだけで評価可能か。\n3. providerの音声機能はmodel/locale/region/API modeでどこまでavailabilityが変わるか。\n4. audio-native/realtime経路とSTT→text LLM→TTS cascadeは同一の運用特性を持つか。"
    },
    {
      "heading": "EVIDENCE TABLE",
      "text": "[E1] official product documentation | Voice Live API Overview | Microsoft | https://learn.microsoft.com/en-us/azure/ai-services/speech-service/voice-live | accessed 2026-09-15 | API family: Voice Live; current docs expose gpt-realtime-1.5/gpt-realtime/gpt-realtime-mini and Data Zone variants | locator: Features / API design and compatibility / model table | VERIFIED_SPEC + PRODUCT_VOLATILE | support/context\n[E2] official API how-to | How to use the Voice Live API | Microsoft | https://learn.microsoft.com/en-us/azure/ai-services/speech-service/voice-live-how-to | accessed 2026-09-15 | api-version=2026-04-10; transcription choices include azure-speech, mai-transcribe(preview), whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe, gpt-4o-transcribe-diarize depending on chat model | locator: WebSocket endpoint / Audio input transcription | VERIFIED_SPEC + PRODUCT_VOLATILE | support/limit\n[E3] official product documentation | What is speech to text? | Microsoft | https://learn.microsoft.com/en-us/azure/ai-services/speech-service/speech-to-text | updated/published 2026-07 approx.; accessed 2026-09-15 | current Speech service | locator: real-time / fast / batch / diarization | VERIFIED_SPEC + PRODUCT_VOLATILE | support\n[E4] official product documentation | Detect different speakers in an audio recording | Google Cloud | https://docs.cloud.google.com/speech-to-text/docs/multiple-voices | accessed 2026-09-15 | Speech-to-Text; V2 request example; diarization_config | locator: Speaker diarization / RecognitionFeatures | VERIFIED_SPEC + PRODUCT_VOLATILE | support/limit\n[E5] official model documentation | Chirp 2: Enhanced multilingual accuracy | Google Cloud | https://docs.cloud.google.com/speech-to-text/docs/models/chirp-2 | accessed 2026-09-15 | model id chirp_2; STT V2; StreamingRecognize language support limited by locale | locator: model details / API methods | VERIFIED_SPEC + PRODUCT_VOLATILE | support/availability\n[E6] official product documentation | Supported voices and languages | Google Cloud Text-to-Speech | https://docs.cloud.google.com/text-to-speech/docs/list-voices-and-types | accessed 2026-09-15 | Chirp 3 HD current voice family | locator: voice type table / Chirp 3 HD notes | VERIFIED_SPEC + PRODUCT_VOLATILE | support/limit\n[E7] official quota documentation | Quotas & limits | Google Cloud Text-to-Speech | https://docs.cloud.google.com/text-to-speech/quotas | accessed 2026-09-15 | current quota table; subject to change | locator: content limits / requests limit | VERIFIED_SPEC + PRODUCT_VOLATILE | limit/operability\n[E8] official product documentation | Partitioning speakers (diarization) | Amazon Transcribe | https://docs.aws.amazon.com/transcribe/latest/dg/diarization.html | accessed 2026-09-15 | current Transcribe | locator: max 30 speakers; batch/streaming speaker labels | VERIFIED_SPEC + PRODUCT_VOLATILE | support/limit\n[E9] official product documentation | Transcribing multi-channel audio | Amazon Transcribe | https://docs.aws.amazon.com/transcribe/latest/dg/channel-id.html | accessed 2026-09-15 | current Transcribe | locator: two-channel identification | VERIFIED_SPEC + PRODUCT_VOLATILE | context/limit\n[E10] official product documentation | Generative voices | Amazon Polly | https://docs.aws.amazon.com/polly/latest/dg/generative-voices.html | accessed 2026-09-15 | Generative engine; regional availability current | locator: region list / real-time and asynchronous synthesis / formats | VERIFIED_SPEC + PRODUCT_VOLATILE | support/availability\n[E11] primary research | VoiceAgentBench: Are Voice Assistants ready for agentic tasks? | Jain et al. | https://arxiv.org/abs/2510.07978 | 2025-10-09; accessed 2026-09-15 | benchmark paper | locator: 5,500+ synthetic spoken queries; tool selection/workflows/multiturn/safety; reported gaps in orchestration, Indic generalization, adversarial robustness | RESEARCH | counter/context\n[E12] primary research | RW-Voice-EQ Bench: A Real World Benchmark for Evaluating Voice AI Systems | Ayllon et al. | https://arxiv.org/abs/2607.14846 | 2026-07-16; accessed 2026-09-15 | benchmark paper | locator: multidimensional TTS/STS/SU/ASR; dimension-specific performance; real-world accent/emotion/noise failures | RESEARCH | counter/context\n[E13] primary research | SEA-SpeechBench: A Large-Scale Multitask Benchmark for Speech Understanding Across Southeast Asia | Liao et al. | https://arxiv.org/abs/2609.09672 | 2026-09-09; accessed 2026-09-15 | 97,194 samples / 99 eval sets / 597h / 11 SEA languages | locator: task and language coverage; temporal/emotion/translation gaps; low-resource prompt gap | RESEARCH | counter/context"
    },
    {
      "heading": "FINDINGS",
      "text": "F1 [VERIFIED_SPEC]: STT、speaker diarization、TTS、realtime voice agentは別能力であり、同じprovider内でも別API/model/modeとして実装される。Microsoft Voice Liveはaudio input/output、noise suppression、echo cancellation、interruption/end-of-turn、function callingを一つのrealtime layerに統合するが、入力transcription modelは別選択である。\nF2 [VERIFIED_SPEC]: diarizationとchannel identificationは同義ではない。Amazon Transcribeはspeaker diarizationで最大30 speakerをlabel可能だが、channel identificationは最大2 channel。Google STTもdiarizationをspeaker change/number labelとして提供する。物理channel分離を話者識別能力へ外挿しない。\nF3 [VERIFIED_SPEC + PRODUCT_VOLATILE]: availabilityはmodel × locale × region × API/modeで変動する。Google Chirp 2はV2専用でStreamingRecognizeのlanguage supportに制約があり、Chirp 3 HD TTSはglobal/eu/us endpointを持つ一方、現行docsではregionalization/data residencyの対象外と明記される。AWS Polly Generative voicesもregion限定。\nF4 [INFERENCE grounded in E1-E10]: realtime voice agentには少なくとも input capture → VAD/turn detection → ASRまたはaudio-native understanding → agent/tool reasoning → response generation → TTS/audio generation → playback → interruption/barge-in のlatency chainがある。単一のend-to-end latency値だけではbottleneckを特定できない。\nF5 [RESEARCH]: VoiceAgentBenchは音声agent評価をtranscriptionからtool selection、multi-tool、multi-turn、安全性へ広げ、実験上orchestration/generalization/adversarial robustnessの不足を報告する。製品保証・failure rateへは外挿しない。\nF6 [RESEARCH]: RW-Voice-EQ BenchはTTS naturalness/expressiveness/identity stability/reliability、STS、speech understanding、ASR robustnessが独立に崩れ得ることを示す。SEA-SpeechBenchも多言語・paralinguistic・temporal理解で大きな差を報告する。したがってWERだけをSpeech AI全体の品質指標にしない。"
    },
    {
      "heading": "LIMITS / COUNTEREVIDENCE / FAILURE ENVELOPE",
      "text": "- clean-speech WER改善はnoise、accent、code-switch、overlap、emotion、far-field、telephony、domain terminologyでの精度を保証しない。\n- diarization speaker IDは通常session内generic labelであり、本人identity verificationではない。speaker recognition/biometricsと混同禁止。\n- TTS naturalnessが高くてもpronunciation、number/date/entity rendering、voice identity stability、latency、interruptibility、安全なconsent運用は別軸。\n- audio-native modelはcascadeより常に低遅延/高品質とは確定しない。network、turn detector、tool latency、synthesis/playbackが支配的になり得る。\n- transcriptが正しくてもtool orchestrationやspoken instruction handlingが失敗し得る。逆にASR文字列差があってもtask successする場合がある。\n- benchmark結果はfixture/language/model snapshot依存であり、特定商用tenantのSLAや現行failure rateを証明しない。"
    },
    {
      "heading": "AVAILABILITY_GATES",
      "text": "provider; model/voice ID; API revision; region/endpoint; locale/language; streaming vs batch; realtime vs async; diarization/channel mode; subscription/resource type; quota/concurrency; preview/GA status; data-residency mode; custom voice/voice cloning consent and eligibility."
    },
    {
      "heading": "IMPLEMENTATION_STATE",
      "text": "Microsoft: Voice Liveはrealtime agent統合層として現行docsで提供。WebSocket api-version 2026-04-10。transcription backend/model選択とchat/realtime model互換性を別管理。\nGoogle: STT V2/Chirp系、speaker diarization、TTS Chirp 3 HD等を別service/modelとして提供。locale/endpoint/feature差を固定して使う必要がある。\nAWS: Transcribeでstreaming/batch、speaker diarization/channel identification、PollyでTTS/generative voice。realtime conversational agent全体はこれら部品を組み合わせる構成として評価し、単一Speech製品とみなさない。"
    },
    {
      "heading": "BENCHMARK_OR_EVAL",
      "text": "Core評価は最低でも次を分離する: ASR WER/CER + entity/number/date accuracy; diarization DER/JERまたはspeaker-attributed word error; TTS intelligibility/pronunciation + naturalness/expressiveness + speaker/voice stability; turn detection false-cut/late-turn; interruption/barge-in success; first-audio latency/turn latency/p95; tool selection/argument/task success; multilingual/code-switch/noise/overlap robustness; safety/privacy behavior。単一総合scoreは補助値に留める。"
    },
    {
      "heading": "SECURITY / PRIVACY / AUDIT / COST / OPERABILITY",
      "text": "Privacy: raw audio、transcript、speaker label、voiceprint/embedding、custom-voice reference、tool transcriptを別artifactとしてretention/consent/access control対象化。diarization labelを本人識別情報と誤認しない一方、voice biometricを扱う場合は別privacy gateを設ける。\nAudit: session ID、audio input hash/locator、model/voice IDs、API revision、locale、timestamps、partial/final transcript、speaker/channel labels、turn boundary、tool calls、interrupt events、final text/audio provenanceを保持。\nCost/Operability: audio duration、characters/tokens、streaming session/concurrency、custom model/voice、network egress等のmeterを分離。quotaはcurrent確認必須。Google TTSのcurrent docsではrequest/content/concurrent streaming limitsが明示され、変更可能性も明記される。"
    },
    {
      "heading": "SAFE_WORDING_FOR_FUTURE_BOOK",
      "text": "「Speech AIはSTTやTTSの単体機能ではなく、認識、話者分離、turn-taking、Agent/tool実行、音声合成、interrupt処理までを含む実時間経路として評価する。各機能のavailabilityと品質はmodel、language、region、API modeで異なるため、provider名だけで能力を一般化しない。WER、自然性、会話latency、task success、安全性は別指標として測る。」"
    },
    {
      "heading": "UPDATE_WORK_HANDOFF",
      "text": "Core候補として SPEECH INTERACTION CONTRACT を追加検討。fields: session_id, provider, region, input_mode, audio_codec/sample_rate, STT/audio-model ID+version, locale, diarization/channel config, VAD/turn config, agent/model ID, tool policy, TTS voice/model ID, output codec, retention/consent, latency trace, transcript provenance, interruption events, task outcome。\nTEST_PROFILE（未実施）: consent済みsynthetic/public audioのみで、clean/noise/far-field/telephony、accent、多言語/code-switch、2-speaker overlap、interrupt、numbers/dates/proper nouns、single/multi-tool taskを固定fixture化。pipeline別に cascade(STT→text agent→TTS) と audio-native/realtime を同一taskで比較し、WERだけでなく task success、DER、turn-cut、barge-in、p50/p95 first-audio/turn latency、cost/minute、audit completenessを記録する。"
    },
    {
      "heading": "OPEN_UNKNOWN",
      "text": "/ STOP_CONDITION"
    },
    {
      "heading": "UNKNOWN",
      "text": "OpenAI direct Realtime APIの2026-09-15時点の正確なcurrent model/API revision・region/data controlsは今回一次資料を十分固定できず、provider mapping採用時に公式docsで再採集する。"
    },
    {
      "heading": "UNKNOWN",
      "text": "provider横断で比較可能な標準化されたend-to-end voice-agent latency/safety benchmarkの成熟度。"
    },
    {
      "heading": "STOP_CONDITION",
      "text": "STT/TTS/diarization/realtime agentの分離、current Microsoft/Google/AWS implementation、availability gates、multidimensional benchmark evidence、failure envelope、future Core contract/test profileまで後続Workが採否判断可能な粒度に到達したためEA-071 bounded packetを閉じる。\n\n---"
    }
  ],
  "human_url": "/tools/ea-071-speech-ai",
  "markdown_url": "/agent/article.md?path=%2Ftools%2Fea-071-speech-ai",
  "json_url": "/agent/article.json?path=%2Ftools%2Fea-071-speech-ai"
}