Azure AI Speech text-to-speech vs Google Cloud Text-to-Speech

Azure AI Speech text-to-speech 80

Read as a model. MAI-Voice-2.1 and MAI-Voice-2.1-Flash entered public preview on 1 October 2026, and Speech SDK 1.52.0 shipped on npm and PyPI on 28 September 2026 (30). Speech release notes for July, August and September 2026. The TTS release notes don't list MAI-Voice-2.1 yet (10). Retirements follow Microsoft's published lifecycle policy with dated notices. The MAI model has turned over three times since March 2026 (MAI-Voice-1, MAI-Voice-2-Flash, MAI-Voice-2.1), and the MAI page dropped MAI-Voice-2-Flash without a dated notice, though the language list still carries it as preview (10). Public release notes and Microsoft Q&A, SDK issue tracker not checked (10 of 25). Speech SDK current in eight languages (15). The SDK is a closed binary, so we can't see its CI (5 of 10).

Google Cloud Text-to-Speech 25

Read as a model. The newest release note is still 30 December 2025 (Chirp 3 HD in yue-HK), 311 days ago, and gemini-3.1-flash-tts-preview appears on the Gemini-TTS page with no dated note (0). In the 90 days from 8 August the Python client shipped 2.38.0 on 1 October, whose only change declares Python 3.15 support, which counts half and can't set the recency line (0). Release notes, a public issue tracker and paid support (10 of 15). Official client libraries, Python at 2.38.0 and Node at 7.2.0 (10 of 15). Package CI not checked (5 of 10). Total 25.

Azure AI Speech text-to-speech 65

Real-time synthesis takes SSML, a W3C format with documented Azure extensions, and we didn't confirm a public OpenAPI file for text-to-speech (10 of 25). No llms.txt found (0). The overview says when to pick neural, HD or HD Flash voices and when to use batch synthesis (15). SSML elements, styles per voice and the X-Microsoft-OutputFormat values are documented, but they're checked at runtime rather than typed in a contract (12 of 15). The REST page lists seven status codes with likely causes, and examples cover REST and the SDKs (13 of 15). Dated release notes and api-version values on batch synthesis (15). The new MAI voices page gives SSML, REST, Python, C#, JavaScript and Java examples for both models. llms.txt per the 1 October check, not re-read on 5 October.

Google Cloud Text-to-Speech 80

Discovery document revision 20260827 with typed methods and 26 schemas, and protocol buffers for gRPC (25). No llms.txt found (0). The Chirp 3 HD and Gemini-TTS pages say which voices suit which use and list formats and limits, but the reference doesn't say when to choose streaming or long audio (15 of 20). Typed messages with enums for encoding and gender and required fields (15). Samples in several languages, standard status codes with little method-specific error detail (10 of 15). Versioned v1 and v1beta1 with dated release notes (15). Total 80.

Azure AI Speech text-to-speech 75

API reading of the checklist. More than 40 output formats chosen by header, from 8 kHz telephony to 48 kHz, and synthesis events in the SDK (23 of 25). The voice list returns the region's voices in one response with no paging documented, while batch jobs list with paging, and per-request limits are published (10 minutes of audio, 50 voice or audio tags, 64 KB per WebSocket turn) (15 of 20). The REST page lists 400, 401, 415, 429, 502 and 503 with likely causes, and the SDK returns cancellation details with error codes (15 of 20). Retry guidance for 429, no idempotency key on batch jobs (10 of 20). Every REST request needs an SSML body and four headers, including the output format and a User-Agent. Speech SDKs in eight languages (12 of 15).

Google Cloud Text-to-Speech 65

API reading of the checklist. Audio returns Base64 in JSON or streams over gRPC, with encoding, sample rate and timepoints selectable (15 of 25). voices.list filters by language code with no paging, and operations list with page tokens (15 of 20). Standard gRPC status codes with messages, generic rather than TTS-specific (15 of 20). Synthesis is stateless and long audio runs as an operation, with no idempotency key (10 of 20). A first call needs a project, billing, OAuth credentials and a voice, and Cloud Storage for long audio. Official client libraries in many languages, of which Python and Node were checked (10 of 15). Total 65.

Azure AI Speech text-to-speech 90

Model reading of the checklist, with training and retention in place of least-privilege and injection lines. Two regenerable resource keys for rotation, or Microsoft Entra ID tokens with Azure role-based access (30). Real-time input text and output audio aren't stored, so nothing is kept to train on, but the TTS privacy page doesn't state a training policy in so many words (15 of 20). Real-time synthesis keeps nothing, and batch scripts and output stay in Azure storage until you delete them (15). Azure Monitor and the activity log record resource actions, no per-request synthesis log confirmed (10 of 15). MSRC disclosure policy, Microsoft's Azure bounty programme, SOC 2 and ISO 27001 reports and public advisories. The microsoft.com security.txt still carries an Expires date of 2026-09-23 on 5 October (20).

Google Cloud Text-to-Speech 80

Model reading of the checklist, with training and retention in place of the least-privilege and injection lines. OAuth 2.0 with service accounts and IAM roles per project (30). The discovery document lists access_token, oauth_token and key as query parameters, so 10 off, as every Google API listing in this run (20 of 30). The data logging page says Google doesn't log any customer Cloud TTS text or audio (20). The service is stateless and keeps no content (15). Because it is stateless, Data Access audit logs don't apply, leaving admin activity only (5 of 15). Valid security.txt, Google's vulnerability reward programme, SOC 2 and ISO 27001 and public Cloud security bulletins, as on google-speech-to-text (20). Total 80.

Azure AI Speech text-to-speech 80

Azure status page with post-incident reviews (20). The 23 July 2026 review (ZJV6-SGG) names Azure Speech in Foundry Tools among services hit by connectivity failures for traffic entering or leaving West US from 14:44 to 19:41 UTC, so we count one major outage. The 1 October note missed it. A 29 September 2026 review covers intermittent 5xx errors for Cognitive Services in Sweden Central from 10:03 to 15:58 UTC, counted as minor. Nothing since 1 October names Speech (10). TTS quotas published, 20 transactions a minute on F0 and 30 a second on S0 by default, adjustable to 1,000, plus batch limits (15). The quotas page explains that 429 usually means backend capacity for a voice and region, and asks for retry logic, a gradual ramp and spreading load across regions (15). Covered by Microsoft's online services SLA (10). Neural and HD voices, the surface graded here, are GA. MAI-Voice-2.1, MAI-Voice-2.1-Flash and MAI-Voice-2-Flash are public preview without an SLA, which doesn't change this line (10).

Google Cloud Text-to-Speech 85

Google Cloud status page with a per-product history (20). None of the four Google Cloud incidents since 8 August (20 August, 1 September, 8 and 9 October, and the europe-west2 incident of 20 October, which listed Speech-to-Text among 38 products) names Text-to-Speech. The public page shows only broad incidents, so clean is read with that limit (30). Quotas published with numbers, 1,000 requests a minute for most voices, 200 for Chirp 3, 500 for Studio, 100 concurrent streams and 100 long-audio requests a minute (15). The quotas page (last updated 26 October 2026) says nothing about the error a breach returns or how to back off, read as google-speech-to-text was (0). A 99.9 per cent monthly SLA with credits of 10, 25 and 50 per cent, read on 6 November (10). Chirp 3 HD and Gemini 2.5 Flash and Pro TTS are GA, with Gemini 3.1 Flash TTS and Flash-Lite in preview (10). 20 + 30 + 15 + 0 + 10 + 10 = 85.