Published by TonePerfect. Vendor facts checked on 10 September 2026 against public pages; no ratings are invented and no vendor paid to be listed.

At a glance

APILanguagesScripted depthUnscriptedEntry priceBest for
SpeechSuper8 (en, zh, de, fr, es, ko, ja, ru)Phoneme / word / sentenceYes (English, IELTS/PTE)$20/mo minimum, $0.004–0.008 per requestBroad edtech coverage, offline SDK
Speechace3 (en, fr, es, with regional variants)Phoneme / syllable / wordYes (Premium)$40/mo for 5,000English exam preparation
Azure AI Speech33 localesPhoneme / syllable / word; prosody on en-USYes, streaming≈ $1 per audio hourBreadth, existing Azure stacks, short clips
ELSA API1 (English)Word / sentence with stress and intonationYesOn requestEnglish consumer apps, exam prediction
TonePerfect7 (zh, en, fr, es, de, it, pt)Sound / word / sentence; Mandarin initial / final / toneNo$10/mo for 2,500Multi-language scripted practice, Mandarin, per-unit decisions

SpeechSuper

The incumbent for language-learning apps. Eight languages, products for words, sentences and paragraphs, semi-scripted and unscripted English (IELTS and PTE style), and an offline iOS/Android SDK licensed per device. Pricing is per request with a $20 monthly minimum and quarterly invoicing; prepaid tiers discount 15–30%. If you need Korean, Japanese or Russian, or on-device scoring, start here. Full comparison →

Speechace

Built around English proficiency: fluency with lexical stress and intonation, IELTS/CEFR/PTE/TOEIC rubric scores on Pro, and spontaneous speech up to two minutes on Premium. Three languages with regional variants (US/UK English, France/Canada French, Spain/Mexico Spanish). Plans bundle requests ($40–125) with $0.008 overage. Best for test-prep products. Full comparison →

Microsoft Azure AI Speech

Pronunciation assessment as a feature of Azure's speech service: 33 locales, accuracy/fluency/completeness scores everywhere, prosody and miscue detail on en-US, unscripted and streaming modes, and per-audio-hour pricing that makes short clips very cheap. The cost is integration: the full feature set lives in the Speech SDK, results need parsing and thresholds are yours to tune. Best for breadth, existing Azure customers and price-sensitive short-clip workloads. Full comparison →

ELSA API

The engine behind a widely used consumer English app, exposed to developers: scripted and unscripted English with stress, intonation and fluency, plus IELTS/TOEFL prediction, trained on a very large non-native English corpus. English only; pricing on request. Best for English-only products that want a recognisable engine. Full comparison →

TonePerfect

Our own API, so judge this paragraph accordingly. Seven languages behind one endpoint with one JSON shape; Mandarin scored as initial, final and tone per syllable with sandhi handled; an explicit per-unit decision so your app knows when to correct and when to stay quiet; word timestamps; audio deleted after scoring; flat plans from $10/month with every dimension included. Scripted speech only, 30 seconds per request, packs labelled experimental while calibration data grows, no free tier. Best for multi-language scripted practice, Mandarin, and teams that want decisions rather than thresholds. Product overview →

How to choose

  1. List your languages. That alone eliminates most options.
  2. Decide scripted vs unscripted. Reading practice and drills are scripted; conversation and exam tasks are not.
  3. Pick the feedback unit. Word colours? Phoneme detail? Mandarin components? Decisions or numbers?
  4. Run the same 20 recordings through your shortlist. A few natives, a few strong learners, a few beginners with known errors. Check who flags natives and who finds the errors.
  5. Model the cost at your volume with the pricing comparison.
  6. Read the data policy. Where does the audio go, and for how long?