BACKGROUND: Understanding speech in noise is a major challenge for most hearing-impaired subjects, with or without hearing aids. To overcome the weaknesses of French-language speech-in-noise tests, we developed a new instrument, with a balanced mix of difficulty of the speech material. METHODS: The speech material comprised 127 sentences taken from the "Marginal Benefit from Acoustic Amplification" (MBAA) corpus, each including 3 keywords. The noise was created using the "onde vocale globale" (global vocal wave: OVG), described by Dodelé. The 127 speech/noise pairs were recorded individually after root-mean-square equalization. The first experiment, on 10 normal-hearing adults, determined the signal-to-noise ratio (SNR) associated with 50% correct keyword identification in each sentence (SNR-50), using an ascending method with noise level set at 73dB SPL. Relative levels between sentences and noise were then adjusted sentence by sentence to achieve an SNR-50 of 0dB. The second experiment, with 12 normal-hearing adults, validated the equalization of sentence difficulty. RESULTS: Mean SNR-50 was -6.64dB (σ=1.47). Mean adjusted SNR-50 was 0.08dB (σ=0.55). Mean psychometric curve slope was 19.3%/dB, with low standard deviations, testifying to the sensitivity of the speech material. CONCLUSIONS: The VRB (vocale rapide dans le bruit: rapid speech in noise) test is based on sentences from the MBAA corpus with background noise based on the OVG at different signal-to-noise ratios. The test is feasible and able to detect slight variations in speech-in-noise performance between subjects.