Analysis

Analysis tools help you move from headline metrics to actual error structure.

`WordErrorAnalyzer`

from plantain2asr import WordErrorAnalyzer

norm >> WordErrorAnalyzer(model_name="GigaAM-v3-e2e-rnnt", top_n=20)

Parameter	Type	Default	Description
`model_name`	`str`	required	Model to inspect
`top_n`	`int`	`20`	Number of top error patterns

from plantain2asr import DiffVisualizer

norm >> DiffVisualizer(model_name="GigaAM-v3-e2e-rnnt", output="reports/diff.html")

This produces a static diff-focused artifact. For broader interactive review, prefer ReportServer or Experiment.save_report_html().

For runtime measurements, prefer the benchmark layer:

benchmarks = experiment.benchmark_models()

This gives latency, throughput, and real-time-factor oriented summaries across supported devices.

All of the following require plantain2asr[analysis]:

Class	Description
`PerformanceAnalyzer`	RTF and latency analysis
`BootstrapAnalyzer`	Bootstrap confidence intervals
`AgreementAnalyzer`	Inter-model agreement
`TopicAnalyzer`	Error by topic or domain
`HallucinationAnalyzer`	Hallucinated word detection
`DurationAnalyzer`	Error vs duration correlation
`NgramErrorAnalyzer`	N-gram error patterns
`CalibrationAnalyzer`	Confidence calibration