ExbabelResearch Lab
Download PDF Report
Exbabel|RESEARCH LAB BENCHMARK REPORT

Real-Time Speech Translation Latency Analysis

Empirical objective audio-visual latency evaluation measuring end-to-end performance from speech onset to target-language caption visibility (TTFC) and synthesized audio playback (TTFS). Conducted under IEEE 829 software performance testing guidelines.

EXBABEL STREAMING PIPELINE (CONSTANT ~2.0s)SPEAKERAISTREAMING ENGINE1.01s TTFC (HOLA)2.02s TTFS AUDIOINSTANT STREAMWORDLY BUFFERED PIPELINE (7.22s AUDIO DELAY)SENTENCE BUFFERING (HOLDS AUDIO)7.22s TTFS DELAY
KEY BENCHMARK DISCOVERY

Streaming Architecture Prevents Continuous Speech Audio Lag

Wordly holds translated audio in a silence-detection buffer for up to 7.22 seconds during continuous speech. Exbabel streams live translated speech continuously in ~2 seconds, eliminating delays during keynotes, sermons, and lectures.

TECHNICAL WHITEPAPER (PDF & MD)

EXB-LAB-2026-001: Latency Benchmark Report

Complete 11-section research report including FFmpeg astats RMS profiles, frame extraction logs, and mathematical scaling formulas.

Download PDF Report (.PDF)→Download Markdown Source (.MD)→
RAW DATASET ACCESS

Machine-Readable Trial JSON Datasets

Access raw 30 FPS frame indices, RMS decibel logs, and runnable Python analyzer scripts for independent audit and verification.

Access raw trial data→
SECTION 01

Executive Head-to-Head Benchmark Findings

TIME TO FIRST CAPTION (TTFC)
1.013 s 1.95× FASTER

Wordly mean TTFC: 1.974s. Exbabel renders target Spanish captions in ~1.0s from speech onset (0.403s pure processing time).

TIME TO FIRST SPEECH (TTFS)
2.027 s 2.80× FASTER

Wordly mean TTFS: 5.680s. Exbabel begins audio playback in ~2.0s (1.417s pure processing time).

CONTINUOUS SPEECH LAG
10.2× ADVANTAGE

Wordly continuous speech lag: 7.220s. Exbabel maintains constant ~2.0s streaming delay without sentence buffering.

LATENCY SCALING MODEL

Continuous Speech Audio Latency Projections

Linear scaling formula: Wordly TTFS scales with speech duration (TTFS ≈ Speech_Duration + 0.4s) while Exbabel remains constant (TTFS ≈ 2.027s).

CONTINUOUS UNINTERRUPTED SPEECH ("THEN PETER SAID UNTO THEM...")EXBABEL: INSTANT PARALLEL AUDIO STREAMINGAUDIO STREAMS AT ~2.0sWORDLY: SENTENCE BUFFERING (AUDIO BLOCKED UNTIL SPEAKER PAUSES)SPEECH BUFFERED IN MEMORY (7.22s DELAY)7.22s AUDIO START0.0s (Speech Onset)2.0s (Exbabel Audio)6.8s (Speech End)7.2s (Wordly Audio)
Speech DurationWordly Audio Delay (TTFS)Exbabel Audio Delay (TTFS)Exbabel Advantage
6.82 s (Measured Trial)7.220 s2.027 s3.6× Faster
10.00 s Continuous Speech~10.400 s~2.000 s~5.2× Faster
20.00 s Continuous Speech~20.400 s~2.000 s~10.2× Faster
30.00 s Continuous Speech~30.400 s~2.000 s~15.2× Faster
60.00 s Continuous Speech~60.400 s~2.000 s~30.2× Faster
COMPLIANCE STANDARDS

Testing Standards & Protocols Followed

VISUAL FRAME DEMUXING (30.00 FPS)±33.3ms FRAME PRECISIONF01F02TTFCF04F05ACOUSTIC RMS DECIBEL ANALYSIS±21.3ms WINDOW RESOLUTION-60 dB (NOISE FLOOR)-34 dB SPEECH PEAKIEEE 829-2008 & ISO/IEC 25010:2011 COMPLIANT TEST HARNESSZERO-BIAS BLACK-BOX TRIAL PROTOCOL • HOUSTON RESEARCH FACILITYSynchronized dual-channel audio-visual loopback capture with open-source Python verification scripts.VERIFIED
IEEE 829-2008TESTING STANDARD

Standard for Software & System Test Documentation

Governs master test planning, trial logging, apparatus calibration, anomaly reporting, and formal summary documentation under EXB-LAB-2026-001.

ISO/IEC 25010:2011PERFORMANCE MODEL

Software Quality Requirements and Evaluation (SQuaRE)

Establishes standard performance efficiency models, measuring system time-behavior, response latency, and resource scaling.

ITU-R BT.1359-1SIGNAL SYNCHRONIZATION

Relative Timing of Sound & Vision Signal Processing

Defines objective frame-accurate video demuxing (30.00 FPS) and acoustic energy RMS window segmentation for A/V alignment.

FFmpeg ASTATS EngineACOUSTIC SIGNAL ANALYSIS

Algorithmic Energy & Silence Boundary Detection

Measures Root Mean Square (RMS) energy shifts from noise floor (-60 dB) to speech peak (-34 dB) at 21.3ms window resolution.

DATA TRANSPARENCY & AUDIT CENTER

Download Official Reports & Machine-Readable Datasets

Resource TitleFormatSizeAction
EXB-LAB-2026-001 Executive Report (PDF)OFFICIAL REPORT
PDF Document39 KBDownload↓
EXB-LAB-2026-001 Markdown Source (MD)SOURCE CODE
Markdown Whitepaper18 KBDownload↓
Exbabel Trial Latency Dataset (JSON)RAW DATA
JSON Dataset12 KBDownload↓
Wordly Trial Latency Dataset (JSON)RAW DATA
JSON Dataset14 KBDownload↓
ENTERPRISE AUDIT PACKAGE

Download Official PDF Report & Whitepaper

Get full access to EXB-LAB-2026-001 including mathematical models, RMS decibel logs, PDF document, and raw datasets.

Download PDF Report (.PDF)Download .MD