A method and system for enhancing dialog determined by an audio input signal. In some embodiments the input signal is a stereo signal, and the system includes an analysis subsystem configured to analyze the stereo signal to generate filter control values, and a filtering subsystem including upmixing circuitry configured to upmix the input signal to generate a speech channel and non-speech channels and a peaking filter configured to filter the speech channel to enhance dialog while being steered by at least one of the control values. The filtering subsystem also includes ducking circuitry for attenuating the non-speech channels while being steered by at least some of the control values, and downmixing circuitry configured to combine outputs of the peaking filter and ducking circuitry to generate a filtered stereo output. In some embodiments, the system is configured to downmix a multichannel input signal to generate a downmixed stereo signal, an analysis subsystem is configured to analyze the downmixed stereo signal to generate filter control values, and a filtering subsystem is configured to generate a dialog-enhanced audio signal in response to the input signal while being steered by at least some of the filter control values. Preferably, the filter control values are generated without use of feedback including by generating power ratios (for pairs of speech and non-speech channels) and preferably also shaping in nonlinear fashion and scaling at least one of the power ratios.