Microsoft Azure Speech Service — Independent Software Review

Transform voice into text and vice versa with advanced AI capabilities.

Compliance Transparency Index

Grade: A — Score: 100/100

Best For

• Large organizations requiring high-volume voice AI solutions within the Microsoft ecosystem.
• Enterprises needing secure and customizable voice integration for their applications.

Not Ideal For

• Small businesses seeking low-cost, simple dictation solutions without technical integration.
• Organizations lacking the technical expertise for significant integration requirements.

Operational Overview

Pricing Structure

Free (F0): $0/month

• Speech-to-Text: 5 audio hours free per month (Standard and Custom; Batch not supported)
• Text-to-Speech: 0.5 million characters free per month (Neural)
• Speech Translation: 5 audio hours free per month
• Speaker Recognition: 10,000 transactions free per month
• Custom Speech: 1 model hosted free per month (unused models decommissioned after 7 days)

Pay-as-You-Go: Usage-based (no minimum commitment)

• STT Standard Real-time: $1/hour; Fast: $0.36/hour; Batch: $0.18/hour
• STT Custom Real-time: $1.20/hour; Batch: $0.225/hour; Training: $10/compute hour
• STT Enhanced add-ons (Language ID, Diarization, Pronunciation Assessment): $0.30/hour/feature (real-time); included for batch
• TTS Neural: $15/1M characters; Neural HD: $22/1M characters
• TTS Custom Professional Voice: $24/1M chars ($48 HD); Training: $52/compute hour; Hosting: $4.04/model/hour
• TTS Personal Voice: $24/1M chars; Profile storage $600/1,000 profiles/month
• Speech Translation Real-time: $2.50/audio hour
• Voice Live Pro: Text $4.40/M tokens; Audio-Standard $17/M tokens; Output $38/M tokens
• Voice Live Standard: Text $0.66/M tokens; Audio-Standard $15/M tokens; Output $33/M tokens
• Voice Live Lite: Text $0.11/M tokens; Audio-Standard $15/M tokens; Output $33/M tokens

Commitment Tiers (Standard): Monthly commitment (volume discounts)

• STT Standard: $1,600/mo for 2,000 hrs ($0.80/hr) to $25,000/mo for 50,000 hrs ($0.50/hr)
• STT Custom: $1,920/mo for 2,000 hrs ($0.96/hr) to $30,000/mo for 50,000 hrs ($0.60/hr)
• TTS Neural: $960/mo for 80M chars ($12/1M) to $15,000/mo for 2,000M chars ($7.50/1M)

Commitment Tiers (Connected Container): Monthly commitment (~5% discount vs Standard)

• STT Standard: $1,520/mo for 2,000 hrs to $23,750/mo for 50,000 hrs
• STT Custom: $1,824/mo for 2,000 hrs to $28,500/mo for 50,000 hrs
• TTS Neural: $912/mo for 80M chars to $14,250/mo for 2,000M chars

Commitment Tiers (Disconnected Container): Annual commitment (for air-gapped/offline deployments)

• STT Standard: $74,100/year for 120,000 hrs to $285,000/year for 600,000 hrs
• STT Custom: $88,920/year for 120,000 hrs to $342,000/year for 600,000 hrs
• TTS Neural: $47,424/year for 4.8B chars to $182,400/year for 24B chars
• Requires sign-up for access

Alternative Consideration

Consider switching to Google Cloud Speech-to-Text: Offers similar functionalities with competitive pricing and integration options.

Frequently Asked Questions

How much does Azure Speech-to-Text cost per hour?

Azure Speech-to-Text pricing depends on the transcription mode. Real-time standard transcription costs $1 per audio hour, fast transcription costs $0.36 per hour, and batch transcription costs $0.18 per hour. Custom models (trained on your domain-specific vocabulary) cost slightly more: $1.20 per hour for real-time and $0.225 per hour for batch, plus $0.0538 per model per hour for endpoint hosting and $10 per compute hour for custom model training. Enhanced add-on features — continuous language identification, diarization (speaker separation), and pronunciation assessment — add $0.30 per hour per feature for real-time use, but are included at no extra charge for batch. All usage is billed in one-second increments.

Is there a free tier for Azure Speech Service?

Yes. The Free (F0) tier provides 5 audio hours per month for speech-to-text (shared between Standard and Custom; batch is not supported), 0.5 million characters per month for neural text-to-speech, 5 audio hours per month for speech translation, and 10,000 transactions per month for speaker recognition (verification, identification, and voice profile storage). You also get 1 custom speech model hosted free per month (unused models are automatically decommissioned after 7 days). The free tier is enough for prototyping and light testing but not for production workloads.

How does Azure Speech pricing compare to Google, Amazon, and Deepgram for transcription?

Azure Speech's standard real-time STT at $1 per hour ($0.0167/minute) is comparable to Google Cloud Speech-to-Text at $0.016/minute. Azure's batch transcription at $0.18 per hour ($0.003/minute) is highly competitive. With commitment tiers, Azure drops to $0.50 per hour for 50,000 hours/month. Deepgram offers rates as low as $0.0043/minute for pre-recorded audio, making it cheaper for pure transcription. Azure's advantage is the breadth of its integrated platform — STT, TTS, translation, Voice Live API, avatars, and speaker recognition all under one API surface with 100+ compliance certifications. Competitors like Deepgram or Soniox may be cheaper per minute but require separate services for translation, TTS, or custom voice.

What is the Voice Live API and how is it priced?

Voice Live API enables end-to-end voice for AI agents by combining speech-to-text, LLM reasoning, and text-to-speech in a single pipeline. It is now integrated with Foundry Agent Service. Pricing is per million tokens across three tiers based on LLM size: Voice Live Pro (GPT-Realtime, GPT-4o, GPT-4.1) starts at $4.40/M tokens for text input and $17/M tokens for audio-standard input; Voice Live Standard (GPT-4o-Mini-Realtime, GPT-4o Mini, GPT-4.1 Mini) starts at $0.66/M tokens for text and $15/M tokens for audio; Voice Live Lite (GPT-4.1 Nano, Phi models) starts at $0.11/M tokens for text and $15/M tokens for audio. Voice Live BYO (bring your own model) provides audio processing only at $12.50/M tokens input and $30/M tokens output.

How much does Azure Text-to-Speech cost and what voice options are available?

Azure TTS offers prebuilt neural voices at $15 per 1M characters, and neural HD voices (higher fidelity) at $22 per 1M characters. Custom professional voice — where you train a unique branded voice — costs $24 per 1M characters for synthesis ($48 for neural HD), $52 per compute hour for voice model training (capped at $936 per training session), and $4.04 per model per hour for endpoint hosting. Personal voice (limited access, requires Microsoft approval) costs $24 per 1M characters with free voice creation and $600 per 1,000 voice profiles per month for storage. Commitment tiers reduce prebuilt neural TTS to as low as $7.50 per 1M characters at the 2,000M character tier ($15,000/month).

What are commitment tiers and when should I use them?

Commitment tiers are monthly volume commitments that reduce per-unit costs by 20–50% compared to pay-as-you-go. For STT Standard, tiers are: $1,600/month for 2,000 hours (effective $0.80/hour, with $0.80/hour overage), $6,500/month for 10,000 hours ($0.65/hour effective), and $25,000/month for 50,000 hours ($0.50/hour effective — a 50% savings over the $1/hour pay-as-you-go rate). For TTS Neural, tiers start at $960/month for 80M characters ($12/1M effective) and go to $15,000/month for 2,000M characters ($7.50/1M effective). You should consider commitment tiers when you consistently process 2,000+ hours of audio or 80M+ TTS characters per month. The risk is that unused committed hours are not refunded.

Can Azure Speech be deployed on-premises or in air-gapped environments?

Yes. Azure Speech supports three deployment models beyond cloud: connected containers run on your infrastructure while maintaining a metered connection to Azure for billing (with ~5% discount over standard commitment pricing — e.g., STT at $1,520/month for 2,000 hours vs $1,600 standard). Disconnected containers operate fully offline for air-gapped environments, priced annually — STT Standard starts at $74,100/year for 120,000 hours, and TTS Neural starts at $47,424/year for 4.8B characters. Disconnected containers require sign-up for access. Embedded Speech provides on-device STT and TTS for scenarios where cloud connectivity is intermittent, using lighter models directly on the device.

What languages does Azure Speech support and what translation options are available?

Azure Speech supports 100+ languages for speech-to-text, with an ever-growing list for speech translation. Real-time speech translation costs $2.50 per audio hour (includes 1 audio input/output and up to 2 text translation languages — additional languages are billed via Azure Translator). Video Translation converts video input at $5 per hour with output at $15 per hour (standard voice) or $20 per hour (personal voice). Live Interpreter enables real-time multilingual communication at $1 per audio input hour plus $10 per 1M characters for text output, $1.50 per hour for standard voice output, or $2 per hour for custom voice output. SDKs are available in C#, C++, Java, Python, and JavaScript.