- tts
- comparison
- azure
- microsoft
- cloud
Microsoft Azure Speech (formerly Bing Speech API) is Microsoft’s cloud TTS platform — 400+ voices, full SSML, custom neural voice training. OfflineTTS runs entirely in your browser on your own device. Here’s how they compare.
Quick Comparison #
| Feature | OfflineTTS | Azure Speech |
|---|---|---|
| Price | Free | Free tier: 500K chars/mo; then $4–$16/1M chars |
| Free Tier | Unlimited (all features) | 500K chars/mo (12 months) |
| Voice Count | 54 + Supertonic/Piper/Kitten | 400+ |
| Languages | 9 | 140+ languages/variants |
| Voice Quality | A/A- (Kokoro 82M) | Neural (excellent), HD (premium) |
| Offline | ✅ After initial model download | ❌ Requires internet |
| Privacy | ✅ Text never leaves your device | ❌ Text sent to Azure servers |
| SSML | ❌ (plain text) | ✅ Full SSML + ExpressAs |
| API | ❌ (browser only) | ✅ REST API, SDKs |
| Signup | ❌ None | ✅ Azure account required |
| Custom Voice | ❌ | ✅ Custom Neural Voice (training) |
| Voice Cloning | ❌ | ✅ (paid training) |
| Export | WAV, MP3 | MP3, WAV, PCM, OGG |
Pricing: Free vs Tiered #
OfflineTTS is completely free. No tiers. No character limits. No expiration.
Azure Speech pricing:
| Model | Price per 1M chars | Free Tier |
|---|---|---|
| Neural | $4 | 500K chars/mo (first 12 months) |
| Neural HD | $16 | Not included in free tier |
| Custom Neural Voice | $16 + training ($200+/hr) | Not included |
The free tier is limited to 500K chars/month and only covers Neural voices (not HD). Post-free-tier, you pay $4/1M for Neural or $16/1M for HD. Custom Neural Voice adds significant training costs on top.
At 10M chars/month:
- OfflineTTS: $0 - Azure Neural: $40 - Azure Neural HD: $160 - Azure Custom Voice: $160+ training
Winner: OfflineTTS. But Azure’s $4/1M Neural tier is competitive with Amazon Polly.
Voice Quality #
Azure Speech offers:
Neural— high-quality, 400+ voices across 140+ languages** Neural HD**— premium enhanced quality, smaller voice set** Custom Neural Voice**— train a branded voice from your own audio
OfflineTTS uses Kokoro 82M (A-grade), Piper, and Supertonic. Kokoro’s top voices are competitive with Azure’s Neural tier.
Where Azure excels:
Custom Neural Voice— unique capability, train a corporate brand voice** 400+ voices**— unmatched variety** Neural HD**— excellent quality for premium voices** Speaking styles**— cheerfulness, sadness, whisper, excitement (via express-as)
Where OfflineTTS excels:
No API latency— instant on-device generation** Privacy**— no data transmitted** Cost**— free regardless of volume
Winner: Azure for voice count and custom voice training. OfflineTTS for cost and privacy.
Privacy & Offline #
OfflineTTS processes everything on your device. Your text is never transmitted. After the initial model download (90–600MB), speech generates without internet access.
Azure Speech processes text on Microsoft’s servers. Your data is subject to Azure’s data handling policies. All requests require internet connectivity. For sensitive content, this means transmitting data to a third-party data center.
Azure offers some compliance certifications (HIPAA, GDPR, SOC 2) that OfflineTTS doesn’t provide. If your organization has specific compliance requirements, Azure’s certifications may matter more than architectural privacy.
Winner: OfflineTTS for architectural privacy. Azure for enterprise compliance certifications. For most individual users, OfflineTTS’s on-device processing is the better deal.
SSML, Styles & Customization #
Azure has the most expressive SSML of any major TTS provider:
<speak version="1.0" xmlns:mstts="http://www.w3.org/2001/mstts">
<!-- Azure's custom express-as tag for emotions -->
<mstts:express-as style="cheerful">
This is incredibly exciting news!
</mstts:express-as>
<mstts:express-as style="sad">
We're sorry to see you go.
</mstts:express-as>
<mstts:express-as style="whisper" styledegree="1.5">
Here's a secret.
</mstts:express-as>
</speak>
Available Azure styles include: cheerful, sad, angry, fearful, excited, friendly, hopeful, shouting, whispering, terrified, unfriendly, cold, embarrassed.
Azure also supports parallel audio streams (<par>
/<media>
) — unique among major providers — for background music + narration in a single SSML document.
OfflineTTS accepts plain text only. No SSML. No expression styles. Pacing is controlled through punctuation only.
Winner: Azure Speech — hands down. No other provider matches Azure’s SSML expressiveness.
Custom Neural Voice #
Azure’s Custom Neural Voice feature lets you train a voice model from recordings of human speakers. This is Azure’s standout differentiator:
Training data: 300–2000+ recorded sentencesTime: Several hours of GPU trainingCost:$200+/hour training + $16/1M chars inference** Quality:Very high for well-recorded data Use case:**Brand voice, celebrity impersonation with permission, character voices
OfflineTTS doesn’t support custom voice training. For voice cloning on your own hardware, see self-hosted F5-TTS or CosyVoice.
Winner: Azure — the only major cloud TTS with production-ready custom voice training.
Pronunciation Assessment #
Azure offers a unique Pronunciation Assessment API that scores speech on accuracy, fluency, completeness, and prosody. This is not a TTS feature — it’s speech evaluation — but it’s part of the same service and valuable for language learning applications.
Winner: Azure (bonus feature not available from any other TTS provider).
The Bottom Line #
Choose OfflineTTS if you:
- Want unlimited free TTS
- Value privacy — text stays on your device
- Need offline capability
- Produce short-to-medium content
- Don’t need SSML or enterprise features
Choose Azure Speech if you:
- Need Custom Neural Voice for brand or character voices
- Require full SSML with emotion styles (express-as)
- Need 400+ voices across 140+ languages
- Are building voice-enabled enterprise applications
- Need pronunciation assessment for language learning
- Require HIPAA, GDPR, or SOC 2 compliance
- Want parallel audio (narration + background) in SSML
Winner for most users: OfflineTTS for personal and small-scale use. Azure Speech for enterprise applications, custom voices, and SSML-heavy production.
Related articles #
Try OfflineTTS #
Four local TTS engines, Whisper transcription, and private browser audio tools.