diff --git a/readersampleprovider.go b/readersampleprovider.go
index 51698c13..77372f67 100644
--- a/readersampleprovider.go
+++ b/readersampleprovider.go
@@ -17,7 +17,6 @@ import (
 const (
 	// defaults to 30 fps
 	defaultH264FrameDuration = 33 * time.Millisecond
-	defaultOpusFrameDuration = 20 * time.Millisecond
 )
 
 // ReaderSampleProvider provides samples by reading from an io.ReadCloser implementation
@@ -41,8 +40,7 @@ type ReaderSampleProvider struct {
 	h264reader *h264reader.H264Reader
 
 	// for ogg
-	oggreader   *oggreader.OggReader
-	lastGranule uint64
+	oggreader *oggreader.OggReader
 }
 
 type ReaderSampleProviderOption func(*ReaderSampleProvider)
@@ -202,18 +200,19 @@ func (p *ReaderSampleProvider) NextSample() (media.Sample, error) {
 		sample.Duration = time.Duration(p.ivfTimebase*float64(delta)*1000) * time.Millisecond
 		p.lastTimestamp = header.Timestamp
 	case webrtc.MimeTypeOpus:
-		pageData, pageHeader, err := p.oggreader.ParseNextPage()
+		// TODO(theomonnom): Read next OpusPacket instead of the next page
+		data, _, err := p.oggreader.ParseNextPage()
 		if err != nil {
 			return sample, err
 		}
-		sampleCount := float64(pageHeader.GranulePosition - p.lastGranule)
-		p.lastGranule = pageHeader.GranulePosition
 
-		sample.Data = pageData
-		sample.Duration = time.Duration((sampleCount/48000)*1000) * time.Millisecond
-		if sample.Duration == 0 {
-			sample.Duration = defaultOpusFrameDuration
+		dur, err := ParseOpusPacketDuration(data)
+		if err != nil {
+			return sample, err
 		}
+
+		sample.Data = data
+		sample.Duration = dur
 	}
 
 	if p.FrameDuration > 0 {
diff --git a/utils.go b/utils.go
index 801f02dd..1d141d0f 100644
--- a/utils.go
+++ b/utils.go
@@ -7,9 +7,60 @@ import (
 	"github.com/pion/webrtc/v3"
 	"github.com/thoas/go-funk"
 
+	"errors"
 	"github.com/livekit/protocol/livekit"
+	"time"
 )
 
+var (
+	ErrInvalidOpusPacket = errors.New("invalid opus packet")
+)
+
+// Parse the duration of a an OpusPacket
+// https://www.rfc-editor.org/rfc/rfc6716#section-3.1
+func ParseOpusPacketDuration(data []byte) (time.Duration, error) {
+	durations := [32]uint64{
+		480, 960, 1920, 2880, // Silk-Only
+		480, 960, 1920, 2880, // Silk-Only
+		480, 960, 1920, 2880, // Silk-Only
+		480, 960, // Hybrid
+		480, 960, // Hybrid
+		120, 240, 480, 960, // Celt-Only
+		120, 240, 480, 960, // Celt-Only
+		120, 240, 480, 960, // Celt-Only
+		120, 240, 480, 960, // Celt-Only
+	}
+
+	if len(data) < 1 {
+		return 0, ErrInvalidOpusPacket
+	}
+
+	toc := data[0]
+	var nframes int
+	switch toc & 3 {
+	case 0:
+		nframes = 1
+	case 1:
+		nframes = 2
+	case 2:
+		nframes = 2
+	case 3:
+		if len(data) < 2 {
+			return 0, ErrInvalidOpusPacket
+		}
+		nframes = int(data[1] & 63)
+	}
+
+	frameDuration := int64(durations[toc>>3])
+	duration := int64(nframes * int(frameDuration))
+	if duration > 5760 { // 120ms
+		return 0, ErrInvalidOpusPacket
+	}
+
+	ms := duration * 1000 / 48000
+	return time.Duration(ms) * time.Millisecond, nil
+}
+
 func ToProtoSessionDescription(sd webrtc.SessionDescription) *livekit.SessionDescription {
 	return &livekit.SessionDescription{
 		Type: sd.Type.String(),