GamiWays

STT / Speech-to-text

Comparison of speech recognition solutions for conversational pipelines (2025–2026). Benchmarks, strategic stakes, and decision questions.

🎯

Strategic framing: read STT in context

What the numbers say

WER describes accuracy on a given corpus; it does not predict proper nouns, accents, noise or turn-taking in your setting. TTFA describes stream start, not conversational responsiveness on its own.

What to combine in the table

For live interaction, read WER and TTFA with streaming, diarization and languages. Endpointing, turn detection and quality on your recordings often matter more than the best isolated score.

Stay alert to

Data residency and retention, stream limits, PII redaction and total cost. The hourly price compares the first public tier; it does not replace load testing or a cloud/self-hosted fallback plan.

Reading method: filter non-negotiable requirements first, then compare accuracy and real-time performance on the same use case, and finally validate with representative audio, languages and noise conditions.

How are the table metrics measured?open

WER: error rate on a specific corpus. A value is comparable only with the same corpus, language, model variant and protocol; it does not directly predict performance on your audio.

TTFA: time to the first transcription fragment for a given stream. “—” means no traceable P50/P95 measurement is published, not zero latency.

RTFx: offline or batch processing throughput: 100× processes one hour of audio in about 36 seconds. It is neither conversational latency nor a single-stream real-time promise.

Streaming chunks: integration settings that define the audio portion and context sent to the model. They must not be read as TTFA. Example: Parakeet’s 2s value is a NeMo chunk, not a latency measurement.

Method: each row gives its source and verification date in the full record. For a decision, compare values on a common protocol first, then test the project’s audio and infrastructure.

Real-time cloud STT APIs — Deepgram Nova-3 is the latency reference (75ms TTFA). Whisper large-v3 is the open-source quality standard. AssemblyAI Universal-3 Pro leads the multilingual WER benchmark — Voice Agent API $4.50/hr (launched Apr 29, 2026).

Click a header to sort

SolutionTTFAWERStreamingMultilingualDiarizationPrice/hr · entrySovereign
Google Gemini 3.5 TranscribeNew
Live STT + recorded-audio analysis — benchmark candidate
——✓85 langs✗$0.54/h✗
Inworld STTUpd
Voice input · evaluation candidate
—✓30 langs✓$0.10/h✗
Deepgram FluxNew
Tour de parole — à mesurer
——✓10 langs✓$0.47/h✓
Speechmatics Melia 1 / Linden 1New
STT français UE — à mesurer
——✓—✓$0.40/h✗
Gladia Solaria-1 / Solaria-3New
STT français UE — live et fichier séparés
——✓100 langs✓API price pending✗
OpenAI gpt-live-transcribeNew
STT live — résidence EEE/CH à contractualiser
——✓—✗$1.02/h✗
Deepgram Nova-3Upd
Phase 1 MVP — Streaming ASR
75ms✓45 langs✓$0.29/h✓
Gradium STTNew
Phase B R&D — Turn detection + real-time infrastructure
100ms✓5 langs✓$0.62/h✓
AssemblyAI Universal-3.6 Pro RealtimeUpd
Voice Agent Pipeline — Accuracy reference
150ms✓32 langs✓$0.21/h✗
ElevenLabs Scribe v2New
STT français — candidat à mesurer
150ms✓90 langs✓$0.39/h✗
Azure Speech (Microsoft)
Swiss sovereignty — institutional
180ms✓100 langs✓$1.00/h✓
xAI Grok STTNew
Phase B R&D — S2S pipeline evaluation
200ms✓25 langs✓$0.10/h✗
Soniox v5New
Accuracy + cost — production validation needed
249ms✓50 langs✓$0.12/h✗
Cartesia Ink-2New
Phase 1 MVP — Full-stack Cartesia pipeline (STT + TTS)
299ms✓5 langs✗$0.54/h✗

Compare by solution

Google Gemini 3.5 TranscribeNew
Live STT + recorded-audio analysis — benchmark candidate
Latencyto measure
0
Accuracy—
0
Price access7/10
7

Google Gemini 3.5 Transcribe separates real-time speech recognition from recorded-audio transcription. Gemini 3.5 Transcribe Live streams interim and final text over the Live API and offers automatic, hybrid or manual VAD plus custom vocabulary. The file model adds speaker diarization and word timestamps, but those two features are not available in the live path. Google announces model-specific WER figures, yet they use a different protocol from the portal’s benchmarks; comparative accuracy and end-to-end latency remain to be measured. Both paths are Gemini Developer API cloud services, not Google Cloud Speech-to-Text v2/Chirp.

Full details →
Inworld STTUpdated
Voice input · evaluation candidate
Latencyto measure
0
Accuracy3.9% WER
7
Price access7/10
7

Realtime STT-1 is a cloud transcription model with 30 documented languages, Voice Profile signals and controllable turn detection. Speaker diarization is experimental. Inference is a separate LLM service; the Realtime API assembles voice components. RAG and embeddings stay application responsibilities.

Full details →
Deepgram FluxNew
Tour de parole — à mesurer
Latencyto measure
0
Accuracy—
0
Price access6/10
6

Flux is a separate Deepgram streaming model from Nova-3. The language overview lists French among 10 languages, with turn detection in the model. On 17 October 2026 multilingual streaming is $0.0078/min ($0.47/h) and English promotional streaming is $0.0065/min. The EU endpoint keeps processing in the EU. No millisecond latency and no Flux AA-WER are published.

Full details →
Speechmatics Melia 1 / Linden 1New
STT français UE — à mesurer
Latencyto measure
0
Accuracy—
0
Price access7/10
7

Melia 1 (17 June 2026) is Speechmatics’ multilingual model, including French and mid-sentence language changes, in batch and realtime. Linden 1 (17 September 2026) is the agent STT: French when language is set to fr, with turn segments that still depend on pauses. Realtime Melia is $0.40/h and Linden $0.30/h. An EU websocket endpoint is documented.

Full details →
Gladia Solaria-1 / Solaria-3New
STT français UE — live et fichier séparés
Latencyto measure
0
Accuracy—
0
Price access6/10
6

Solaria-1 is Gladia’s live model, with 100+ languages including French. Solaria-3 (10 June 2026) covers recorded European audio (EN, FR, DE, ES, IT) and is not a streaming turn model. Self-serve prices on 17 October 2026 are €0.68/h live and €0.16/h for Solaria-3 up to 100 hours. The DPA names OVH in Gravelines, France.

Full details →
OpenAI gpt-live-transcribeNew
STT live — résidence EEE/CH à contractualiser
Latencyto measure
0
Accuracy—
0
Price access5/10
5

gpt-live-transcribe is OpenAI’s live transcription model, separate from gpt-realtime-2.1 and from GPT-Live-1. The price list is $0.017 per minute of realtime audio. French is a language hint. Server VAD is not on this model. The data-controls page documents EEA and Switzerland processing on eu.api.openai.com when account conditions are met. gpt-4o-transcribe is deprecated, with removal planned for 26 February 2027.

Full details →
Deepgram Nova-3Updated
Phase 1 MVP — Streaming ASR
Latency75ms
10
Accuracy5.2% WER
8
Price access7/10
7

Deepgram Nova-3 is a real-time ASR option for voice agents, with 45+ language support, built-in VAD and endpointing. The Voice Agent API provides a complete STT+LLM+TTS pipeline in a single WebSocket. On 17 October 2026 the standard Voice Agent rate is $0.075/min; the September promotional rate has ended. Streaming Nova-3 Monolingual is $0.0048/min on the same date. Deepgram's TTS family includes Aura-2 ($0.030/1k characters) and Flux TTS; Flux is also a separate conversational STT. Enterprise on-premise deployment and an EU endpoint are available. The 75ms P90 value remains a historical Pipecat benchmark, not a live Deepgram SLA.

Full details →
Gradium STTNew
Phase B R&D — Turn detection + real-time infrastructure
Latency100ms
9
Accuracy6.8% WER
9
Price access9/10
9

Gradium STT features semantic VAD for intelligent turn detection (meaning-based, not just silence). Best-in-class accuracy claimed by founders who invented neural audio codecs. 5 languages (FR/EN/DE/ES/PT), word-level timestamps, speaker diarization. Native LiveKit and Pipecat integration. On-premise Enterprise option with zero data retention.

Full details →
AssemblyAI Universal-3.6 Pro RealtimeUpdated
Voice Agent Pipeline — Accuracy reference
Latency150ms
7
Accuracy3.1% WER
10
Price access5/10
5

On 29 September 2026 AssemblyAI released Universal-3.6 Pro Realtime (`universal-3-6-pro`): 32 languages, $0.45/hr, and entity-aware endpointing. Its English voice-agent WER of 5.19% is a vendor protocol and is not the sheet’s comparative WER. Earlier context: AssemblyAI launched its Voice Agent API on April 29, 2026: a complete STT+LLM+TTS pipeline in one WebSocket connection at a flat $4.50/hr. Universal-3 Pro Streaming (u3-rt-pro) is its real-time STT model, with semantic and acoustic turn detection, native barge-in, and session resumption. JSON Schema tool calling supports a custom RAG stack (Pinecone, LlamaIndex, and others) through function calling; there is no native RAG, but external integration is complete. There is no native voice cloning: voices are predefined (18+ English and multilingual voices), though an external TTS such as ElevenLabs or Cartesia can provide a custom voice. LeMUR features cover transcription summarisation, Q&A and sentiment. Supports 99 languages, diarisation and word-level timestamps.

Full details →
ElevenLabs Scribe v2New
STT français — candidat à mesurer
Latency150ms
8
Accuracy2.2% WER
9
Price access7/10
7

Scribe v2 is ElevenLabs’ current speech-to-text pair: batch at $0.22/hour and realtime at $0.39/hour, both listed on the API pricing page on 17 October 2026. French is in the vendor’s excellent-accuracy language tier. The public Artificial Analysis table gives Scribe v2 an AA-WER of 2.2%. The 150 ms figure is the vendor’s realtime claim, not an Où est Ava ? measurement.

Full details →
Azure Speech (Microsoft)
Swiss sovereignty — institutional
Latency180ms
7
Accuracy5.9% WER
9
Price access3/10
3

Azure Speech offers enterprise-grade STT with 100+ languages, custom model training, and Swiss data center (Zurich). 5.9% WER on English. Disconnected container deployment for partial sovereignty. Most expensive option but strongest enterprise compliance. Swiss German custom model available.

Full details →
xAI Grok STTNew
Phase B R&D — S2S pipeline evaluation
Latency200ms
7
Accuracy6.9% WER
8
Price access9/10
9

Grok STT is the standalone transcription API built on the same stack as Grok Voice Think Fast 2.0. It achieves 6.9% WER globally across phone calls, meetings, video/podcasts and telephony — outperforming ElevenLabs Scribe v2 (9.0%), Deepgram Nova-3 (11.0%) and AssemblyAI (12.9%) in xAI's internal benchmark. Particularly strong on entity recognition (phone numbers, dates, currencies) thanks to advanced Inverse Text Normalization. Supports 25+ languages, word-level timestamps, speaker diarization, multichannel audio, and keyterm biasing. Priced at $0.10/h batch — among the most competitive in the market. Best evaluated as part of the Grok Voice Think Fast 2.0 S2S pipeline ($0.08/min all-in).

Full details →
Soniox v5New
Accuracy + cost — production validation needed
Latency249ms
8
Accuracy3.8% WER
10
Price access9/10
9

Soniox v5 claims the best WER on the market at 1.25% (internal benchmark). Priced at $0.12/hr — the most affordable STT API in the benchmark. 249ms TTFS, 50+ languages, speaker diarization. On-premise available on Enterprise. Very new — production track record limited. New entry June 2026.

Full details →
Cartesia Ink-2New
Phase 1 MVP — Full-stack Cartesia pipeline (STT + TTS)
Latency299ms
7
Accuracy1.47% WER
10
Price access7/10
7

Cartesia Ink-2 still holds the June 2026 Pipecat English WER of 1.47%. On 17 September 2026 the docs mark ink-2 stable in English, French, Hindi, Japanese and Spanish. The only hour price on the 17 October pricing FAQ is $0.39 on the Scale plan. The June latency is not a French Où est Ava ? measurement.

Full details →